Ressourcenverzeichnis
Finden Sie, was Sie für Ihren Workflow benötigen.
Guides: techniques, setup, and troubleshootingDurchsuchen Sie Sweet Tea, Pipes, externe Workflows, Modelle, Knotenpakete und Anleitungen in einem faktenbasierten Katalog.
Beginnen Sie stattdessen mit einem Workflow ↗Suchergebnisse
Alle Ressourcen
20 angezeigt of 42.552
Depth-guided image generation with Flux
Generate a new Flux image from a prompt using the estimated depth and spatial layout of an uploaded source image. For a first run, upload a source image, replace the example prompt with a description of your desired subject and appearance, and leave the source defaults in place: 1 megapixel, depth LoRA strength 1, prompt guidance 10, and 20 generation steps. Seed -1 randomizes generation; use a fixed nonnegative seed when comparing settings. The selected output is the final generated image, not a separate depth-map result. The source is resized while retaining its aspect ratio, then Lotus estimates its depth. The inverted depth map conditions Flux.1 dev with its fixed depth LoRA. Output dimensions follow the resized source, subject to latent encoding. Changing output megapixels or resize interpolation also changes the input to depth estimation. Advanced controls separate depth LoRA strength, Flux prompt guidance and final generation sampling. Generation denoise acts on the conditioning-created latent, so it does not directly preserve source colors or texture. Sampler CFG is independent of Flux prompt guidance. Negative conditioning is zeroed and there is no separate negative prompt. Depth estimation has its own sampler, steps, scheduler and denoise controls. Its source configuration uses Euler, one step, normal scheduling, full denoise and disabled added noise; a downstream stage fixes the first sigma at 10000. These settings affect the estimated depth used by generation. Depth estimation and the depth LoRA remain integral stages; there is no bypass or alternative generation mode. The workflow uses the configured Flux.1 dev, Flux depth LoRA, CLIP-L/T5 encoders, Flux VAE, Lotus depth model and Lotus VAE. Source RGB supplies depth guidance rather than direct appearance copying, and uploaded alpha transparency is not carried through to the result.
LTX-2.3: Text to video with audio
Create one video with generated audio from a scene prompt using LTX-2.3. Describe the scene, actions over time, camera movement, visual details, and desired sounds or dialogue. This graph accepts no image, video or audio uploads. For a first run, replace the example scene prompt and retain the defaults: requested size 1280 × 720, duration 5 seconds, 25 FPS, Enhance prompt enabled, Text-only mode enabled, and automatic output format and codec. The workflow generates video and audio together at half the requested spatial dimensions, applies fixed 2× video latent upscaling, then refines both and returns the combined result. Refinement runs every time. Enhance prompt chooses a generated rewrite as the video conditioning text. Disable it to encode your original wording directly. Advanced enhancement controls include sampled or unsampled text generation, token budget, sampling filters, template selection and text encoder LoRA strength. Temperature and the enhancement seed are relevant to sampled rewriting. Thinking mode is available only if the loaded text model supports it. Width and height request the final spatial size; latent alignment may affect realized dimensions. The initial latent receives half each dimension and has a source minimum of 64 pixels with a 32-pixel step. For aligned custom sizes, use final dimensions of at least 128 pixels in multiples of 64. Swap width and height for portrait framing. The retained 1280 × 720 default is preserved; arbitrary dimensions are not guaranteed to produce an exact requested size. Duration is positive whole seconds. Requested frame count is duration × FPS + 1. Use 1–120 FPS and keep the requested frame count at or below the audio latent's source limit of 1000. Video latent length has an eight-frame step, so frame alignment can affect the realized duration; for aligned custom timing, make duration × FPS a multiple of eight. Frame rate also controls video conditioning, generated audio timing and final playback. Text-only mode bypasses solid-color first-frame conditioning in both stages. Turn it off to use the chosen solid color, independently adjustable initial and refinement conditioning strengths, and conditioning compression. Color is a packed RGB integer: 0 is black and 16777215 is white. This mode uses a generated color image rather than an uploaded reference. Initial generation and refinement have independent seeds, CFG values, samplers and comma-separated noise schedules. The distilled LoRA strength affects both stages. Keep the supplied LoRA strength, Euler samplers and schedules for the first run. Negative conditioning is supplied to both stages, but ordinary classifier-free negative guidance has no effect at their default CFG of 1; increasing a stage's CFG makes its negative conditioning relevant. Generation, refinement and sampled prompt enhancement have separate seeds. Use -1 for fresh randomization, or fix every relevant seed when comparing settings. Advanced output controls select the video container, codec and encoding CRF. Keep their automatic settings and default CRF for the first run; the supplied evidence does not establish compatibility for every explicit container and codec pairing.
MiniMax H3 video with audio
Generate a widescreen video with accompanying audio from a text description using MiniMax H3. Describe the scene, action, visual style, camera movement, dialogue, sound effects, and music in the Scene and audio prompt. There is no image upload or separate negative prompt in this Pipe. For a first run, replace the example with a short scene suitable for the two-second duration default and retain the sampling defaults: 20 steps, res_multistep, simple scheduler, and denoise 1. The saved example describes five seconds; shorten its storyboard or increase Duration before using it. Prompt timestamps and camera directions describe your intent rather than guaranteeing exact cuts or timing. Duration is requested in seconds. At fixed 24 fps, the workflow rounds seconds × 24, applies a minimum of five frames, then snaps upward to a valid 17k+5 frame count. Two seconds produces 56 frames, about 2.33 seconds, so the delivered clip can be longer than requested. Output size is not adjustable here: the retained source resolution selector uses its saved 16:9, 0.4-megapixel settings. Advanced controls change generation: Seed -1 chooses fresh random starting noise each run; a fixed seed lets you compare changes with the same starting noise. Steps sets the sampling step count, Sampler selects the sampling method, Scheduler distributes noise levels across those steps, and Denoise selects the fraction of the noise schedule used. Lower denoise does not preserve an uploaded image because this Pipe has none. Video frames and audio are decoded from the same sampled result and delivered together as one video at 24 fps. Use the default 8-bit output for compatibility. The supported 10-bit option can retain smoother gradients but requires compatible encoding and playback. The configured MiniMax H3 diffusion model, Qwen3VL text encoder, video VAE, and audio VAE must be available in the execution environment. This Pipe does not expose model selection, image conditioning, end frames, or an audio-only output.
Stable Audio 3 audio creator
Create an MP3 audio clip from a written description using Stable Audio 3 Medium. No uploaded audio is required. Describe the sound source, instruments or materials, mood, rhythm, and production character you want. For a first run, keep the tropical-house example, Music category, Expand prompt enabled, and Duration at 150 seconds. Keep the source sampling settings: 8 steps, CFG 1, LCM sampler, simple scheduler, and Denoise 1. Run the tool and listen to the MP3 result. MP3 quality changes encoding and file size rather than the generated performance. Prompt expansion uses Qwen to elaborate your description. Category must be exactly Music, Instrument, SFX, or One-shot. Music requests an instrumental track; Instrument requests instrument-focused material; SFX describes effects or ambience; One-shot requests a short isolated sample. These are prompting modes, with no stem extraction or guaranteed seamless looping. Category does not change Duration automatically: set a shorter duration yourself for short effects or samples. Turn Expand prompt off to send your original description directly to audio conditioning. Category and expansion tuning then do not affect the selected conditioning. This selection does not promise that the runtime skips loading or executing every expansion dependency. Duration controls the empty audio latent in seconds. The connected latent node supports 1–1000 seconds. Use whole seconds because the expansion request converts Duration to an integer. Category templates contain their own length suggestions, so expanded wording can conflict with your requested duration; the latent duration remains controlled by Duration. Disable expansion when you want direct control over prompt wording. Text requests do not guarantee exact BPM or event timing. Audio seed and Prompt seed control separate random stages. Use -1 for fresh randomization. To compare audio settings with consistent wording, disable expansion or fix Prompt seed and retain the same expansion settings. Prompt sampling off disables text sampling; temperature and token-sampling controls apply to sampled expansion. Expansion length limit is measured in text tokens, not audio seconds. Thinking mode depends on model support, and generated text has no separate reasoning-text cleanup stage before audio conditioning. Negative prompt describes unwanted sounds. At the default CFG of 1, it does not contribute a positive-versus-negative guidance difference. Higher CFG changes that guidance behavior. Retain Denoise 1 for the first run because generation starts from an empty latent. Audio batch controls the number of latents generated together; larger batches require more memory. The runtime must provide the fixed source models: stable_audio_3_medium.safetensors in checkpoints and t5gemma_b_b_ul2.safetensors in text_encoders, supplied by https://huggingface.co/Comfy-Org/stable-audio-3; and qwen3.5_2b_bf16.safetensors in text_encoders, supplied by https://huggingface.co/Comfy-Org/Qwen3.5. These source dependencies are retained even though the supplied catalog has not resolved them.
Renaissance Style
A Textual Inversion Style Embedding trained on a beautiful late-medieval town archway scene. You can expect to get chaotic red brick/stone and wooden buildings, with floral accents, and a golden hue in your images by...
Wan 2.7 image to video
Animate an uploaded first-frame image into a 2–15 second video with prompted motion and automatically generated background music or sound effects. The uploaded image determines the video's aspect ratio. For a first run, upload your starting image, replace the example motion and scene prompt to match it, and keep 720P resolution and the 10-second duration. Describe the action, camera movement and atmosphere you want. The main prompt supports English and Chinese. The negative prompt describes unwanted visual features; its supplied Chinese default discourages artifacts, distorted anatomy, static scenes and unwanted text. Choose 720P or 1080P and a duration from 2 to 15 seconds in whole-second increments. Prompt enhancement is enabled by default and uses AI assistance to expand your instructions; disable it to submit your wording directly. Leave seed at -1 for randomization on each run, or enter a nonnegative seed to control generation variation. Enable watermark if you want an AI-generated watermark on the result. Under advanced video output settings, leave file type and codec at auto for automatic selection, or choose a container and video codec explicitly. This tool uses a remote API and requires configured provider access. It accepts one starting image. End-frame and audio uploads are not connected, so this tool does not offer an uploaded ending frame or uploaded audio to drive generation.
Wan 2.2 image to video
Generate one silent video from a starting image and a motion prompt using Wan 2.2's paired 14B image-to-video models. For a first run, upload an image, replace the example prompt with the subject's action and camera movement, and keep 640 × 640, 5 seconds, 16 FPS, batch size 1, and Fast mode off. Choose width and height in multiples of 16 to suit the image's composition. An upload is needed while the Source image stage is enabled. Disable that stage to generate without starting-image conditioning, and describe the subject and scene in the prompt as well as their motion. This mode retains the same image-to-video model pair. Uploaded transparency is not preserved by the connected RGB image output. Fast mode selects both Lightx2v acceleration LoRAs and the fast sampling controls. Defaults change from 20 steps, CFG 3.5, and handoff step 10 to 4 steps, CFG 1, and handoff step 2. Only the selected mode's steps, guidance, and handoff apply. Keep total steps between 1 and 10000, guidance between 0 and 100, and each handoff greater than zero and below its total steps. The two LoRA strengths apply only in fast mode. Both stages have independent sampler, scheduler, and noise-schedule shift controls; the supplied recipe uses matching Euler samplers, simple schedulers, and shifts of 5. Duration controls generation length. The graph requests floor(duration × FPS + 1) frames and uses the same FPS for playback. Use a positive duration and 1–120 FPS; choose duration × FPS as a multiple of four for the model's 4n + 1 frame convention. Defaults request 81 frames at 16 FPS, approximately five seconds. Increasing resolution, duration, or FPS increases generation work. Keep batch size 1 for one clip; larger batches still pass through a single video-assembly branch rather than separate clip outputs. Use seed -1 for a new random seed each run or a fixed seed to compare changes with the same initial noise. Advanced controls include a negative prompt for unwanted traits; its influence depends on guidance. The tool returns one video through Sweet Tea, using automatic container and codec selection by default. Available containers are mp4, mkv, and webm; explicit codecs are h264 and av1. Keep automatic selection unless you need a particular compatible combination. Eight-bit output is the default; 10-bit output requires compatible encoding and playback. The graph has no audio input, end-frame conditioning, or upscaling stage.
Z-Image realistic image generator
Write a scene prompt. The realism-focused negative prompt is editable; CFG and seed are also available. Leave seed at -1 to randomize it. Output size is fixed at 1280 × 1920.
All-in-One. Z-Image, Ernie, Klein, t2i, i2i, controlNet, inpaint, outpaint, LLM, WAN, Qwen image, SDXL, Chroma — v6.3
!!Please update Timesaver and RES6LYF nodes!!
Glorious Style
This is a DreamArtist Textual Inversion Style Embedding trained on a single image of glorious, imposing castles, towers, and ivy covered ruins. You can get stunning backdrops in your images by calling Style-Glorious...
Runway image generation
Generate images with Runway from a text prompt and an optional reference image. For a first run, leave Reference image disabled, replace the example portrait prompt with your desired subject, composition, lighting, and style, and keep Output format at 1024:1024 for a square image. Run the Pipe to view the generated images. To use reference guidance, enable Reference image and upload one image. The upload feeds the generator's optional reference-image input; keep the prompt focused on the desired result. Disable Reference image to return to text-only generation. This workflow exposes no reference-strength or exact-editing control. Output format controls the generation format. Use a value supported by the Runway node; 1024:1024 is the supplied square default. Generation uses the Runway partner API and requires configured API access and sufficient account credits. The prompt and enabled reference image are sent to the provider. Generated images are returned for preview through Sweet Tea's output transport; this graph does not permanently save them through SaveImage.
Flux Kontext image editing
Produce one image with Flux.1 Kontext dev using an uploaded image and written instructions. Use it for subject changes, background edits, style changes, or a new composition inspired by the reference. For a first run, upload a Source image, leave Second reference disabled, and replace the supplied swan portrait prompt with your own Edit instructions. Describe the desired change and anything that should remain unchanged. For example: “Change the background to a garden while keeping the subject's pose and clothing unchanged.” Keep the source defaults of Flux guidance 2.5, Denoise 1 and 20 Steps initially. For two references, enable Second reference and upload another image. The images are joined into one canvas before scaling and encoding. Explain each image's intended role in the instructions. In Reference layout, choose the second image's position, whether to match its adjoining dimension to the source, and an optional gap in pixels. Spacing color matters only when the gap is greater than zero. These layout controls apply only when the second reference is enabled. Output dimensions follow the reference canvas after automatic Kontext scaling; there are no independent width or height controls. Adding a second reference or changing its layout can change the output proportions. The encoded canvas supplies both the starting image and reference conditioning. Denoise ranges from 0 to 1: the default 1 allows full regeneration with reference conditioning, while lower values retain more of the starting canvas and can constrain the requested changes. Seed -1 selects a fresh random seed each run. Use a fixed seed to compare settings with the same starting noise. Steps controls the number of sampling steps; Sampler chooses the sampling algorithm and Scheduler chooses its noise schedule. The source defaults are 20, euler and simple. Flux guidance controls the model's prompt guidance. Sampling also exposes an independent Classifier-free guidance scale, retained at its source default of 1. There is no editable negative prompt; negative conditioning is zeroed internally. Uploaded transparency is not preserved because only the image loaders' RGB outputs are connected. Requests to preserve details are instructions to the model, rather than masked protection of particular pixels.
Magnific precise image upscale
Enlarge and enhance an uploaded image with Magnific Precise V2, adjusting edge definition, texture and fine detail. For a first run, upload your source image and keep the defaults: 2x enlargement, photo style, sharpening 7, smart grain 7, fine detail 30, and automatic input downscaling enabled. Run the Pipe to receive the enhanced image. Choose 2x, 4x, 8x or 16x enlargement. Use sublime for general images, photo for photographs, or photo_denoiser for noisy photographs. Sharpening controls edge definition, smart grain adds grain and texture to counter excessive smoothness, and fine detail controls added textures and micro-detail. Each enhancement control ranges from 0 to 100. Added texture and detail can change the source's appearance. Automatic input downscaling reduces the input when the requested output would exceed the provider's maximum resolution. The result may therefore be smaller than the selected factor implies relative to your original upload. Disabling this option does not remove the resolution limit; the supplied source does not specify that limit or the behavior when it is exceeded with downscaling disabled. This Pipe returns the enhanced image only. Uploaded transparency is not preserved through this graph: the image loader's separate alpha mask output is not connected.
Arthemy Anima comic image
Generate a single comic-style image or a batch from text with Arthemy Comics Anima v2. No uploaded image is required. The supplied prompt describes an elf brawler with gradients, hatched shadows and a dramatic camera angle; edit it to choose your subject, composition, background and visual treatment. For a first run, keep the supplied positive and negative prompts, 1024 × 1344 portrait dimensions and image count of 1, then run the Pipe to receive generated images. The negative prompt discourages unwanted features and styles, including photo, 3D, watercolor and flat-color treatments. Revise both prompts when changing the intended style. Width and height are in pixels, in increments of eight; together they determine the aspect ratio. Image count selects the batch size. Larger dimensions and batches require more memory. Advanced Sampling contains seed, steps, CFG, sampler, scheduler and denoise. Seed -1 randomizes each run; use a fixed seed to compare changes with the same initial noise. The source defaults are 35 steps, CFG 5.5, er_sde sampling and the simple scheduler. CFG controls prompt guidance; excessive guidance can harm the result. Sampler and scheduler choices are supplied by the runtime and change the sampling process. Keep denoise at 1 for full sampling from the empty latent; reducing it changes the sampling range and does not provide image-to-image editing. Runtime requires Arthemy_Comics_Anima_v2.safetensors, Arthemy_Qwen3_6b.safetensors and qwen_image_vae.safetensors. These assets are fixed in this Pipe, and their availability must be established in the runtime. The source's model-merging, block tuning, restoration, chaos and LoRA examples are disconnected from this output. This Pipe generates images and does not expose those operations or export model weights.
Grok text to image
Generate images from a text prompt with Grok and preview the results in Sweet Tea. No source image is required. For a first run, use the supplied neon arcade portrait prompt with grok-imagine-image, square aspect ratio, 1K resolution, and one image. Replace the prompt to describe your own subject, composition, lighting, and style. Choose an aspect ratio for the image's width-to-height shape, select the provider's 1K or 2K resolution tier, and request 1–10 images per run. The model selector offers grok-imagine-image, grok-imagine-image-quality, and grok-imagine-image-pro. Seed controls whether the provider node reruns; it does not make generated images reproducible. Leave it at -1 to randomize that value on each run. This workflow has one main prompt and no separate negative prompt. Results are returned for preview; the workflow does not permanently save image files.
Danger Zoo ~ Johnsons Negative Embeddings ~ Helper Collection!
I put this together to make it easy for people to get them all in one page. Full details and examples can be found on the original model pages.
Gemini image refinement
Refine or edit an uploaded image with Gemini, or generate an image from a prompt without a reference. For a first editing run, enable Source image, upload an image and keep the supplied refinement prompt, Auto aspect ratio and 4K resolution. Adjust the prompt to describe the changes and details to retain. The supplied prompt asks for sharper detail while retaining composition and text; these are model instructions, not guarantees of exact preservation. For text-only generation, disable Source image and replace the refinement prompt with a description of the complete image you want. An upload is required while Source image is enabled. Auto aspect ratio follows the reference image when supplied; choose an explicit ratio when you need a particular canvas shape, especially without a reference. Resolution selects 1K, 2K or 4K; the source describes native Gemini upscaling for 2K and 4K. Model selects the Gemini image model for editing or generation. Seed -1 randomizes each run; a fixed seed requests best-effort consistency, but identical results are not guaranteed. Provider response format requests IMAGE or IMAGE+TEXT, but this Pipe returns only the generated image and discards accompanying text. System instructions set foundational model behavior; the supplied instructions prioritize producing an image and interpreting abstract input visually. Generation sends the prompt and any enabled reference image to Gemini through a paid provider call. Uploaded transparency is not carried through the connected RGB image path. The selected output is an image preview; this graph does not save a local image file.
FireRed image edit
Edit an uploaded image with text instructions using FireRed Image Edit 1.1 and return one edited image. For a first run, upload your source image, replace the example positive prompt with the changes or finished appearance you want, and leave output size at 1 megapixel with Lightning mode off. The resized source supplies both visual conditioning and the initial latent. Output size preserves its aspect ratio. Denoise defaults to 1 for a full transformation; lower values retain more source structure while the image remains available as visual conditioning. Lightning mode selects the supplied FireRed Image Edit 1.0 Lightning 8-step LoRA together with separate sampling settings. Its defaults are 8 steps and CFG 1; standard mode uses 40 steps and CFG 4. Only the active mode's step and CFG values affect the result. Lightning LoRA strength affects the selected model only when Lightning mode is on. Negative prompting is optional and has little or no influence at CFG 1. Use seed -1 for a fresh random seed, or a fixed seed to compare settings with the same initial noise. Sampler and scheduler affect either mode; Euler and Simple are the source defaults. Resampling method controls how the source is resized before conditioning and encoding. Advanced guidance normalization and sampling shift also apply in both modes; retain their source defaults for the first run. Step values must be integers from 1 to 10,000, and CFG values must stay between 0 and 100, even though the upstream value controls expose broader entry ranges. This tool requires one source image and does not provide mask-based editing or preserve uploaded transparency. The base model, text encoder, VAE and Lightning LoRA file are fixed to the source configuration. The edited image is returned through Sweet Tea's preview output; this graph does not permanently save an image file.
HiDream E1.1 image editor
Edit an uploaded image using natural-language instructions with HiDream E1.1. Upload a source image and describe the change you want, including details you want retained. The Pipe resizes the source to a target pixel count, uses it as image conditioning, and returns an edited image. For a first run, upload your image and replace the example hair-edit instruction with a request suited to it. Keep the default 1 megapixel resolution, 20 steps, Euler sampler, simple scheduler and guidance settings. Resolution approximately follows the source aspect ratio. Resize method controls interpolation of the source before editing. There is no mask input, so requests to leave other regions unchanged are instructions rather than guarantees. Uploaded transparency is not preserved. In the default regular guidance mode, instruction guidance compares the edit-prompt prediction against the negative-prompt prediction, with image conditioning in both branches. Secondary guidance compares the negative-prompt image-conditioned prediction against the edit-prompt text-only baseline. These controls are coupled; change one at a time. Guidance combination also offers nested mode, which changes how the two comparisons are combined. Guidance normalization provides a strength control. Normalize before CFG is off by default, retaining post-CFG attenuation. Turning it on rescales before CFG without clamping and can amplify predictions; it interacts with normalization strength and both guidance values. Advanced sampling controls expose steps, sampler, scheduler and denoise. Denoise changes the noise schedule and is not a direct source-pixel preservation control. Seed -1 produces a new random variation each run; use a fixed seed when comparing settings. This Pipe produces a single edited image and has no mask-based editing or alpha-preservation mode.
SDXL Turbo
Generate images from text with the fixed SDXL Turbo checkpoint. Describe your subject, setting, composition, and visual style in Positive prompt; no uploaded image is required. The workflow requires sd_xl_turbo_1.0_fp16.safetensors in ComfyUI's checkpoints directory. For a first run, replace the example positive prompt and keep 512 × 512 pixels, one image, one sampling step, CFG 1, Euler ancestral, Denoise 1, and Add noise enabled. Width and height use pixels in multiples of 8 and determine the aspect ratio. Image count generates a batch of variations. Larger dimensions and batches increase memory use. In advanced Sampling, seed -1 selects new random starting noise each run; use a fixed seed to compare changes from the same starting noise. Advanced controls expose steps, sampler, CFG, denoise, and Add noise. Steps range from 1 to 10; additional steps increase sampling work without guaranteeing a better result. Changing the sampler changes generation behavior. The negative prompt has no effect at CFG 1; raising CFG introduces its influence and changes the source Turbo preset's behavior. Denoise changes the noise schedule used with the empty latent, not the strength of editing an input image. Keep Add noise enabled for normal generation; disabling it removes starting random noise without supplying a source image. This tool produces generated images and does not offer image editing or checkpoint selection.
