Trouvez ce dont vous avez besoin pour votre flux de travail.

Photo to geometry maps

Upload one perspective photo to produce two geometry-map images from a shared MoGe-2 estimate. For a first run, upload your image, keep the defaults, and run the Pipe. The first image shows OpenGL normals and the second shows DirectX normals, with inference detail set to 9, automatic camera field of view, and invalid-region filtering enabled. Choose each map independently: OpenGL normals, DirectX normals, depth, colored depth, or the predicted validity mask. OpenGL normals use green = +Y; DirectX normals use green = −Y. Both outputs use the same geometry estimate and inference settings. In Geometry estimation, inference detail ranges from 0 (fastest) to 9 (most detail), as described by the source. Enter the camera's horizontal field of view in degrees when known, or leave 0 to recover it automatically from predicted points. This controls the focal length used to unproject depth into 3D. Invalid-region filtering marks predicted sky and invalid pixels as invalid in points and depth; disable it to retain raw predictions. The predicted mask remains selectable either way. Uploaded transparency is not used as a mask. Automatic resizing checks image width: images wider than 2048 pixels are resized to a 2048-pixel longer edge. Images whose width is at most 2048 remain unchanged, including tall images. Disable automatic resizing to use the original image size. Inference batch size specifies images per inference call, rather than output variations; this Pipe supplies one photo. Force projection is enabled by default, but the supplied source does not explain its effect on the maps. These images are estimates from visible photo content, not guaranteed accurate measurements. This Pipe returns image maps and does not export a mesh or GLB. It requires the fixed moge_2_vitl_normal_fp16.safetensors checkpoint under models/geometry_estimation/.

Flux Kontext image editor

Create one edited image with Flux.1 Kontext Pro using an uploaded image and written instructions. Upload your source image, keep Second image disabled, and try the supplied pop art instruction with the default square output. Describe the subject or region to change and anything you want preserved. The Pipe returns one generated image. To combine two images as editing context, enable Second image and upload another image. Explain in your instruction how each image should be used. Reference layout places the second image relative to the source, matches their adjoining dimensions, and sets the gap and its color. These settings apply only when Second image is enabled. Both images are stitched into a single model input; they are not separate model reference sockets. No second upload is needed while Second image is disabled. Output aspect ratio sets the generated image shape independently of the reference layout. Enter a width:height ratio between 1:4 and 4:1. Guidance controls prompt guidance strength, and Steps sets the number of generation steps; the supplied defaults are 3 and 50. Seed -1 selects a new random seed each run; enter a fixed seed to compare settings. Prompt expansion allows the provider to modify the instruction for more creative generation and makes results nondeterministic even with a fixed seed. This workflow uses a partner API and requires configured access and sufficient credits. Uploaded transparency is not preserved: only the image loaders' RGB image outputs are connected, not their alpha masks.

FLUX1024x10248GB VRAM

LTX-Video 2.31024x10248GB VRAM

LTX 2.3 talking and singing portraits

Create talking or singing portrait videos from uploaded audio and portrait images with LTX 2.3. The workflow generates successive segments, spatially upscales and refines them, blends their boundaries, and saves individual segment videos plus an assembled H.264 MP4 at 25 FPS. Both saved outputs use the selected original recording, including accompaniment. For a first run, upload speech or singing and supply an image in all four portrait slots. Reuse the primary portrait in unused slots because every image loader remains connected. Replace shot 1's example prompt, "hello", with a visual description of the subject, setting, expression and motion. Prompts describe visuals; the recording supplies the voice. The source's intended single-shot setup leaves prompts 2–4 empty. Filling them supplies additional image/prompt pairs to the shot looper, but the supplied evidence does not establish its exact empty-slot filtering or rotation order. Audio start time is measured in seconds and defaults to 30; set it to 0 to begin at the start. Duration 0 loads the remaining recording. Select a short duration for an initial trial. Chunk duration is a positive target in whole seconds, not total video duration. One chunker determines loop count from the original recording; another derives segment timings from separated vocals. Both receive the same target duration, but identical boundaries or counts are not established. Their independent variation controls retain the source value of 10; units and exact effects are unspecified. Width and height control shared crop framing for shots 1 and 2, with separate crop positions. Base generation dimensions come from shot 1's crop at half scale, followed by a two-times spatial upscaler; rounding can affect final dimensions. Shots 3 and 4 skip that initial crop, so prepare compatible framing. Only RGB image outputs are connected, so uploaded transparency is not preserved. Generation and refinement have independent image strengths and disable switches. Disabling one stage's image guidance leaves the other stage unchanged and does not remove any upload requirement. Portrait compression is applied before both conditioning stages. The static-camera LoRA affects both sampling stages; strength 0 removes its contribution. Three additional slots accept names of compatible installed LoRAs, each with its own strength. None leaves a slot inactive. Generation steps, schedule shifts, stretching and terminal sigma control the initial sampling schedule. Refinement uses its separate comma-separated sigma sequence. Each stage has an independent seed, CFG value and sampler. Seed -1 randomizes that stage on each run; fix both seeds when comparing other settings. The negative prompt is shared and its influence depends on each stage's CFG. Keep the source sampling defaults for the first run. Decode tiling, overlap and last-frame handling are separate advanced controls; their exact conventions are not fully established by the supplied evidence. Dissolve frames controls boundary blending: 20 frames equals 0.8 seconds at 25 FPS. Overlap trimming is calculated separately, so retain the transition defaults initially. The dissolve curve is a string field; use only values supported by the installed transition node. Final and segment CRF settings independently control H.264 compression. Lower CRF means less compression and larger files. Encoded segments are reloaded for assembly, so final encoding cannot recover detail lost during segment encoding. Final encoding trims the video to the selected audio. Segment outputs are produced within the generation loop and cannot run independently. Longer recordings require more total processing and storage. The workflow relies on its configured models and custom nodes; the supplied dependency catalog does not resolve every required asset.

Wan 2.2 controlled video

Generate a prompted video with Wan 2.2 Fun Control 5B, guided by a source video's frames or extracted edges and an optional reference image. For a first run, upload a source video, replace the example prompt with your desired scene and motion, and either upload a reference image or disable the Reference image stage. That stage is enabled initially, so leaving its upload empty requires disabling it. Keep 704 × 704, 121 frames, 24 FPS, batch size 1, and the default sampling settings. Canny edge guidance is disabled initially: original source frames supply guidance. Enable it to use extracted edges instead. Its normalized low and high thresholds apply only in that mode; keep the low threshold below the high threshold. Reference image guidance can be disabled independently. Generation uses these inputs as conditioning rather than directly editing the source video's pixels. Width and height are pixels in multiples of 16. Frame count uses 4n + 1 values, such as 81 or 121. Playback FPS is independent of generated frame count: 121 frames at 24 FPS yield approximately 5.04 seconds. Changing FPS changes playback duration without requesting more frames. Source audio and source FPS are not carried into the output. Advanced controls include unwanted traits in the negative prompt, seed, sampling steps, prompt guidance, sampler, scheduler, denoise fraction, and sampling shift. Seed -1 randomizes each run; a fixed nonnegative seed reuses sampling noise. Denoise is not a direct source-video preservation strength. The retained negative prompt is the source's Chinese-language list of unwanted traits and can be edited. Video playback offers 8-bit or 10-bit creation; 10-bit support depends on downstream encoding and players. Video output offers container and codec choices; leave both on auto initially. The graph returns one video output with no separate edge preview. Batch size changes the latent batch but does not provide separate output branches for individual clips, so keep it at 1 for an ordinary single-clip run.

Wan 2.21024x10248GB VRAM

Gemini image edit and generation

Edit a reference image or generate images from text using Gemini. For a first run, upload a source image, keep the Source image stage enabled, and describe the changes you want and the features to retain. The supplied example requests anime styling while retaining identity. Keep the defaults: Gemini 3 Pro Image Preview, Auto aspect ratio, 1K resolution and seed -1 for randomization. For text-only generation, disable the Source image stage and replace the editing instruction with a complete scene description. An upload is required while the source stage is enabled. Auto aspect ratio follows the reference when supplied; choose an explicit ratio for text-only generation when output shape matters. Resolution selects a 1K, 2K or 4K target. The source uses Gemini's native upscaler for 2K and 4K. Model selects between Gemini 3 Pro Image Preview and Nano Banana 2 (Gemini 3.1 Flash Image). A fixed nonnegative seed offers best-effort repeatability, but identical results are not guaranteed, including when model or request settings change. System instructions set foundational behavior alongside your image instruction. The supplied instructions direct the model to produce images and interpret requests visually. Model response mode selects IMAGE or IMAGE+TEXT; this Pipe returns images only, so any accompanying text is not surfaced. This tool connects one reference image and has no mask-based editing input or multi-reference batching stage. Uploaded alpha transparency is not preserved because the loader's separate alpha-mask output is not connected.

Wan 2.2 controlled video · Four-step

Generate a silent video from a scene prompt and control footage using Wan 2.2 Fun Control with a four-step Lightx2v preset. For a first run, upload a control video and replace the example scene prompt. Leave Canny edge extraction disabled for already prepared control frames. For ordinary footage, enable Canny edge extraction; its normalized thresholds apply only when enabled, and the low threshold should stay below the high threshold. Reference-image conditioning is disabled initially. To use it, upload an image and enable Reference image. Leave it disabled when no image is supplied. The reference upload supplies RGB image conditioning; the loader's alpha mask is not connected. Start with 640 × 640 pixels, 81 frames and 16 playback FPS, producing approximately 5.1 seconds. Width and height use increments of 16 pixels; frame count follows 1 + 4n. Playback FPS is independent of source FPS. Source audio is not included, and there is no source-FPS resampling control. The result is newly generated footage guided by control frames. Leave batch size at 1 for a single clip; this Pipe exposes one video output rather than separate outputs for batch items. Both generation stages run for every result. Initial sampling uses the high-noise model for steps 0–2; refinement uses the low-noise model for steps 2–4 without adding fresh noise. Both use a four-step schedule, Euler sampling, the simple scheduler, CFG 1, sampling shift approximately 8 and their corresponding Lightx2v LoRA at strength 1. Refinement is part of generation rather than a standalone postprocessing mode. The standard sibling uses a twenty-step path without this LoRA pair. Use seed -1 for randomization or a fixed nonnegative seed to reuse initial noise with unchanged inputs and settings. Advanced controls expose independent guidance, samplers, schedulers, schedule boundaries, model shifts and LoRA strengths. Preserve the supplied preset initially. When changing schedules, keep both schedule lengths equal, coordinate schedulers, match Initial end step to Refinement start step within the schedule, and set Refinement end step to complete it. LoRA strength changes do not automatically change the schedule. Negative conditioning is available but may have no effect at CFG 1. Leave output container, codec and CRF on automatic selection initially. CRF -1 uses automatic selection; explicit values range from 0 to 63 and their effects depend on the codec. Explicit container and codec choices must be compatible. Video construction offers 8-bit and 10-bit representation; the source documents smoother gradients with 10-bit and possible player or downstream compatibility limitations.

Wan 2.21024x10248GB VRAM

LTX-Video 2.31024x10248GB VRAM

Wan 2.5 text to image

Generate still images from English or Chinese prompts with the Wan 2.5 text-to-image preview API. The resulting image appears in the Pipe preview. For a first run, replace the example prompt with a description of your subject, composition, lighting, and style. Keep the default 1024 × 1024 dimensions and prompt enhancement enabled, then run the Pipe. No uploaded image is required. Optionally add a negative prompt describing elements or visual characteristics to avoid. Width and height are independently adjustable from 768 to 1440 pixels in increments of 32; together they determine the aspect ratio. Prompt enhancement allows the service to expand your description. Disable it to submit your wording without automatic expansion. Add watermark requests an AI-generated watermark on the result. Seed -1 randomizes each run; a nonnegative seed requests a particular randomization. The model is fixed to wan2.5-t2i-preview. This tool produces still images even when the example or your prompt describes camera movement. Generation sends your prompt to a remote partner API and requires configured ComfyUI API access, connectivity, and sufficient account credits.

Edit video with Grok

Edit a short uploaded video with written instructions using Grok Imagine. For a first run, upload a clip no longer than 8.7 seconds and no larger than 50 MB. Replace the example motorcycle prompt with the changes you want for your clip, such as subjects, action or visual style. Leave Seed at -1, then run to receive one edited video in Sweet Tea. This workflow uses grok-imagine-video and requires a source video. The clip and prompt are sent to the Grok provider through a paid provider call. Changing Seed triggers a fresh provider run; reusing a fixed seed does not guarantee identical results. The edited video output defaults to automatic file type and codec selection. The output section also supports MP4, MKV or WebM file types and H.264 or AV1 codecs. These settings control output encoding, while the editing prompt controls the requested creative changes.

LTX 2.3 audio-driven lip-sync video

Generate an audio-driven speaking or singing video from a reference image and a selected segment of uploaded audio, with image-aware prompt expansion, 2× spatial upscaling and refinement. Output defaults to an H.264 MP4 containing the cropped uploaded recording. For a first run, upload an image with a clearly visible face and a speech or singing recording. Set Crop start and Crop end in minutes:seconds within the recording, with the end after the start. The defaults select 0:33–0:48; adjust both for your upload. Replace the performance description to match your subject, expression, movement and camera behavior, explicitly mentioning speaking or singing with the recording. The source description requests a woman smiling and singing toward a fixed camera. Review the negative prompt, whose unrelated scene details and exclusions of smiling and looking at the camera can conflict with your intent. Prompt expansion requires the configured Gemini provider and uses the original uploaded image and performance description to produce the positive video prompt. Camera directions such as static, dolly in/out, orbit and focus shift are interpreted through that prompt; they are not deterministic camera controls. The cropped recording conditions both video sampling passes and supplies the exported soundtrack, rather than generated audio. Start with a 1280-pixel longest edge, original proportions and 25 FPS. Output dimensions are calculated and rounded to a multiple of eight before initial generation at half dimensions, followed by 2× spatial upscale and refinement. Changing proportions affects the separate reference resize: Center crops to fit, while Disabled stretches. The sizing stage's letterbox image and background color are not used as the generation reference. Custom ratio components apply only in a custom-ratio mode supported by the installed sizing node; the supplied evidence does not establish its exact selector value. Conditioning image size separately sets the reference preprocessing longest edge, with a default of 1536 pixels, before compression. Use positive dimensions and positive integer FPS. FPS drives frame count, model conditioning and export playback together. Frame count is cropped duration converted to integer seconds, multiplied by FPS, plus one. Fractional durations can therefore cause timing differences; use whole-second crop boundaries for a first run and inspect synchronization. Exact lip alignment and subject fidelity are not guaranteed. Uploaded transparency is not preserved. Initial and refinement image guidance have independent strengths and disable switches. Disabling a pass's image conditioning leaves that sampling pass active; the image remains required for dimensions and Gemini context. Upscaling and refinement still run. The optional fixed style LoRA affects both passes and starts enabled: bypass Style LoRA or set its strength to zero to remove its influence. Its filename suggests a xianxia style, but the supplied evidence does not establish its appearance. The separate distillation adapter also affects both passes; begin with its source strength of 0.5 and four initial steps. Advanced controls include separate seeds, prompt guidance and samplers for each pass. Seed -1 randomizes each run. Fixing video seeds does not freeze the Gemini-expanded prompt, and a fixed prompt-expansion seed does not establish deterministic provider responses. Prompt variation controls provider temperature. Steps, Scheduler and Denoise affect initial generation only; refinement uses its explicit comma-separated noise levels ending at zero. Ordinary negative guidance has no effect at CFG 1. Keep MP4/H.264, CRF 19, yuv420p, 8-bit and sRGB for the source encoding preset. Alternative file types, codecs, pixel formats and bit depths require compatible output transport and encoder support. GIF cannot retain the soundtrack. HDR output settings alone do not establish HDR generation. Use the shared FPS control to change playback timing. The configured provider, source models and custom nodes are required. Active loaders request gemma_3_12B_it_fp8_e4m3fn.safetensors, ltx-2.3-22b-dev.safetensors, ltx-2.3-spatial-upscaler-x2-1.1.safetensors and the two fixed LoRAs. Dependency metadata contains differing filenames; use the active loader choices.

ModèleCivitai

Aissist XL - Negative Embedding

AissistXL is a negative embedding fine-tuned to be working on Animagine XL v3 and its derivatives. It trained on the standard negative prompt for Animagine XL v3 plus some extra parameters to make sure you always...

TextualInversionSDXL 1.0

ACE-Step1024x10248GB VRAM

Text to sound effects

Generate an MP3 sound effect from a text description using ElevenLabs eleven_sfx_v2. For a first run, replace the traffic example in Sound description with the effect you want, or keep it to try the supplied example. Keep the 5-second duration, looping disabled and prompt influence at 0.3, then run the tool. No uploaded audio is required. Duration accepts 0.5–30 seconds. Enable Seamless loop to request a smoothly looping effect. Prompt influence ranges from 0 to 1; higher values make generation follow your description more closely. Generation audio format selects the generator's MP3 or Opus response. The connected output stage saves either response as MP3, so this choice does not change the final file type. In the advanced MP3 output section, choose V0, 128k or 320k for the final encoding setting. The tool provides one connected MP3 output.

ACE-Step 1.5 music generator

Generate music with ACE-Step 1.5 XL from a musical description and lyrics, and export MP3. No media upload is required. For a first run, keep the supplied description, lyrics, and defaults, then generate: the example requests a two-minute ambient song with English vocals at 72 BPM in E minor. Musical instructions guide generation rather than guaranteeing exact performance. To create your own track, describe genre, instruments, mood, vocal delivery, production, and arrangement in Music description. Enter sung words and optional bracketed section cues in Lyrics, then select the matching vocal language. Keep tempo, key, and meter references consistent across the text and musical controls. The time-signature selector offers source values 2, 3, 4, and 6, with no separate denominator control. Output duration sets the audio length in seconds. Conditioning duration is independent: update both to the same value when changing length. Track count sets the batch size using the same musical conditioning. Seed is shared by audio-code generation and diffusion sampling; -1 randomizes it, while a fixed nonnegative integer lets you reuse a seed with the same settings. Generate audio codes enables the additional language-model stage. Its guidance, temperature, top-p, top-k, and min-p controls apply only when enabled. Disabling it leaves the text-conditioned diffusion path available. Advanced Audio sampling controls set diffusion steps, CFG, sampler, scheduler, and denoising strength. Sampling configuration exposes the source sampling shift, defaulting to 3. Full denoising defaults to 1 because generation starts from an empty audio latent. Negative conditioning is automatically zeroed; there is no editable negative prompt. MP3 quality selects V0 variable-bitrate encoding or fixed 128k or 320k encoding and affects export rather than music generation. The workflow uses the ACE-Step XL base diffusion model, its two Qwen text encoders, and the ACE 1.5 audio VAE. It has no uploaded audio reference, continuation, or audio-editing input.

Hunyuan3D1024x10248GB VRAM

Flux de travailCivitai

AUTO-SEND FILES PRIVATE/SECURE 🔐: ComfyUI Output Folder >> Encrypt >> Send to GoogleDrive / Dropbox 😎👍 — v1.0.0

Sick of manually downloading your ComfyUI output files from the cloud? Want them to just show up in your GoogleDrive automatically? (or DropBox - Link Below)

Kling image generation

Generate images from text with Kling Image O1 or Kling 3.0 Omni and view the results in Sweet Tea. For a first run, keep Kling Image O1, 1K resolution, square aspect ratio, and image series disabled. Use the supplied fashion portrait prompt or replace it with your own subject, composition, lighting, and style. Describe unwanted elements in the same prompt when needed; there is no separate negative prompt. Choose the aspect ratio for image proportions and 1K, 2K, or 4K resolution for the requested output size. To request a series of 2–9 images, select Kling 3.0 Omni and then choose the image series count. Keep image series disabled with Kling Image O1, which does not support that mode. This Pipe accepts text only; no reference image input is connected. Leave seed at -1 to randomize the run seed. Seed controls whether generation re-runs, rather than reproducibility: results remain non-deterministic even with a fixed seed. Generation sends the prompt to the paid Kling provider. Results are delivered through Sweet Tea's image output; the workflow does not permanently save image files.

[LuisaP] Sci-fi/Hard surface

Needs the keyword " structure " ex: a cyberpunk structure , a building structure , a turret structure .

Hunyuan3D texture from images

Generate a texture image by reconstructing an object with Hunyuan3D through the provider API. For a first run, upload a front view of the object and leave Back view bypassed. Keep model 3.1, 500,000 faces and PBR enabled, then run the Pipe. To include a back view, enable Back view and upload a matching image of the same object. Model version offers Hunyuan3D 3.1 or 3.0. Face count requests between 3,000 and 1,500,000 mesh faces; it does not set texture pixel resolution. PBR requests physically based material generation, but this Pipe returns only the texture image. Seed -1 requests a fresh run. The provider produces non-deterministic results even when a seed is reused; the seed controls node re-running and caching. Generation is fixed to Normal for the texture output. Geometry mode does not supply this deliverable, and LowPoly requires a polygon-type control absent from this interface and is unavailable with model 3.1. This Pipe runs the underlying 3D generation on each requested generation run. It does not accept an existing mesh. It delivers no GLB or OBJ model, rendered model preview, or separate metallic, normal or roughness maps.

Wan 2.2 four-step frame-to-video

Generate a silent video from a scene and motion prompt, optionally guided by a start image, an end image, or both. For a first run, upload a start image and replace the sleeping-cat prompt with your subject, action, camera movement and style. Upload an end image if you want ending-frame guidance; otherwise disable the End image stage. Both image stages initially remain enabled, so upload an image for each enabled stage or disable it explicitly. Leave the default 640 × 640 dimensions, 81 frames, 16 FPS and sampling settings for the first run, then generate and view the returned video. The default playback lasts approximately 5.1 seconds. The image stages can be disabled independently: start only animates from an opening image, end only provides ending guidance, both guide a transition, and disabling both leaves prompt-driven generation. Images provide guidance rather than a guarantee of an exact transition. Only RGB image outputs are connected; uploaded alpha masks are unused. There is no audio input in this tool. Width and height are pixels in increments of 16. Frame count follows 1, 5, 9 and so on and changes how many frames are generated. Playback FPS changes duration without generating additional frames: duration is frame count divided by FPS. Keep batch size at 1 for a single video; this Pipe has one video output rather than separate outputs per batch item. Advanced settings preserve separate initial and finishing-stage controls. Seed -1 randomizes initial noise; a fixed nonnegative seed repeats the noise choice with the same other settings. The high-noise stage adds noise and runs steps 0–2, then the low-noise stage continues steps 2–4 without adding fresh noise. Both use a total schedule length of 4, CFG 1, Euler, simple scheduling and sampling shift 8. The fixed high/low-noise Lightx2v LoRAs each have model strength 1. If deliberately changing this recipe, coordinate both total step counts and the segment boundaries; changing one field alone can interrupt the handoff. LoRA strengths do not select the standard 20-step sibling recipe. The supplied negative prompt is a Chinese list of unwanted visual defects and content; retain it or edit it for your scene. Output container, codec, CRF and bit depth control video representation and encoding. Leave container and codec on auto and CRF at -1 for the first run. Ten-bit output may be unsupported by some players or downstream nodes.

Wan 2.21024x10248GB VRAM

Qwen-Image-Edit1024x10248GB VRAM

Action figure image edit

Create an articulated toy or action-figure style image from an uploaded image. Describe the edit in the positive prompt using "action the [thing]", such as "action the scene"; optionally add exclusions in the negative prompt. Fast mode selects the Lightning LoRA branch with four sampling steps by default; disabling it selects standard sampling. Advanced controls adjust edit strength, LoRA strengths, seed, sampler, scheduler, and each mode's steps and guidance.

Product placement video

Create a product-placement video from three product photos and one composition reference. Nano Banana generates a new start frame from all four images; Gemini examines it and writes a motion prompt; Kling animates it. Every run includes composition generation. This tool does not accept an existing finished start frame or return the intermediate still as a separate output. For a first run, upload all four required images: products are images 1–3, and the composition reference is image 4. Edit Placement prompt to describe how the products should appear in that scene. Its default assumes products held in a hand in a fanned arrangement; revise it for other compositions. In Motion instructions, describe desired subject movement and camera motion. These instructions go to Gemini, whose response becomes Kling's actual prompt automatically. Keep the initial 1K start frame, 5-second duration, 1080p video, and automatic output encoding settings. Start frame aspect ratio determines the composition shape and the resulting video's framing. Auto follows an input image's aspect ratio, but the supplied source does not identify which of the four takes precedence. Choose an explicit ratio when framing matters. Kling's own aspect-ratio field is ignored with the generated start frame connected. Start frame resolution and Video resolution are independent. The three model selectors act independently. Image model offers Nano Banana 2 (Gemini 3.1 Flash Image) and Nano Banana 2 Lite. Both retain all four image bindings and share aspect-ratio and thinking-level controls. For Start frame resolution, enter 1K, 2K, or 4K with the full model; set 1K before switching to Lite, which supports only 1K. Thinking level offers MINIMAL or HIGH. Image temperature controls randomness from 0 to 2, and Image nucleus sampling controls the sampling threshold from 0 to 1; lower values favor more focused choices. Keep their supplied defaults initially. Image generation instructions changes foundational behavior; use Placement prompt for ordinary scene changes and preserve image production. Motion writer model offers gemini-2.5-pro, gemini-2.5-flash, gemini-3-pro-preview, gemini-3-1-pro, and gemini-3-1-flash-lite. Motion writer instructions optionally changes its foundational behavior. Leave that field empty initially; any resulting response is consumed directly as the video prompt. Video model offers kling-v3 and kling-3.0-turbo. kling-v3 supports 4k, 1080p, and 720p; Turbo supports 1080p and 720p, so switch away from 4k before selecting Turbo. Generate audio controls kling-v3. Turbo always generates native audio and ignores this toggle. The retained mode generates a single shot lasting 3–15 seconds; storyboard controls are not exposed. Negative prompt describes unwanted elements for video generation and does not affect the still composition. A seed of -1 randomizes each run. Fixed image and motion-writer seeds request best-effort repeatability without guaranteeing identical results. Kling's seed controls re-running rather than reproducible video generation. File type and Codec affect the returned video's encoding; keep auto initially. Inspect product identity, packaging, label text, pose, and motion before using the result: prompt requests are not guarantees of exact reproduction. The upload loaders' alpha-mask outputs are not connected, so uploaded transparency is not preserved through a separate mask path. Text and thinking-image outputs from image generation are not returned. For a still composition without motion writing or animation, use the separate Product placement photo tool.

1024x10248GB VRAM