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 41.745
Qwen-Image: Text to image
Generate images from text with Qwen-Image, using standard sampling or an optional Lightning LoRA mode. No source image is required. For a first run, replace the example positive prompt with your subject, composition, lighting and style. Include desired written text in the prompt. Leave the negative prompt empty unless you want to discourage particular elements. Keep 1328 × 1328 pixels, image count 1 and Lightning off. Standard mode defaults to 20 steps and CFG 4. Lightning mode selects Qwen-Image-Lightning-8steps-V1.0.safetensors and its separate defaults of 8 steps and CFG 1. The toggle switches the model, steps and guidance together. Each mode has independently editable steps and CFG; only the active mode's settings affect the result. Lightning strength applies only in Lightning mode. Negative-prompt influence depends on CFG; Lightning's default CFG is 1. Width and height are pixels in multiples of 16. Set them together to choose the aspect ratio; 1664 × 928 is landscape and 928 × 1664 is portrait. Increasing dimensions or image count increases memory use. Seed -1 selects a fresh random seed each run; use a fixed seed when comparing settings. Sampler, scheduler, denoise and sampling shift are shared by both modes. The supplied defaults are Euler, simple, full denoising at 1 and sampling shift approximately 3.1. This graph starts from an empty latent and does not edit uploaded images. The fixed model files must be available: qwen_image_fp8_e4m3fn.safetensors in diffusion_models, qwen_2.5_vl_7b_fp8_scaled.safetensors in text_encoders and qwen_image_vae.safetensors in vae. Their source is https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/tree/main/split_files. Lightning additionally uses Qwen-Image-Lightning-8steps-V1.0.safetensors in loras, supplied at https://huggingface.co/lightx2v/Qwen-Image-Lightning/resolve/main/Qwen-Image-Lightning-8steps-V1.0.safetensors. Dependency evidence currently marks these files as missing; ensure they are available before running the corresponding setup.
HiDream I1 Full text to image
Generate images from text using the fixed HiDream I1 Full FP8 model. No uploaded image is required. Describe the subject, composition, lighting and style in the positive prompt; use the negative prompt for unwanted attributes. For a first run, replace the example positive prompt or run it as supplied. Keep 1024 × 1024 pixels, image count 1 and the supplied advanced settings: 50 steps, CFG 5, uni_pc sampler, simple scheduler, sampling shift 3 and denoise 1. Run the Pipe to receive the generated image batch. Width and height must be multiples of 16; set them together to choose the aspect ratio. Image count controls how many images are generated per run. Increasing dimensions or image count increases resource use. Seed -1 randomizes each run; use a fixed seed to reuse the initial noise when comparing settings. Advanced sampling exposes steps, CFG, sampler, scheduler, denoise and sampling shift. CFG controls prompt guidance; at CFG 1 the negative prompt has no effect. Sampler and scheduler change the sampling process, while sampling shift adjusts the model's noise schedule. Denoise defaults to 1; lowering it reduces the effective noise schedule. Because generation starts from an empty latent, it does not preserve an existing image. The model and text encoders are fixed. Dev and Fast models are not selectable modes, and this Pipe does not offer image-to-image input.
Chroma1-HD text to image
Generate an image or batch of images from text using the fixed Chroma1-HD model. No uploaded image is required. For a first run, replace the positive prompt with your subject, composition, lighting and style. The example requests a tiger photograph with CHROMA1-HD title lettering. Review the negative prompt: its default discourages sketches and cartoon aesthetics, so remove those exclusions when you want those styles. Begin with 1024 × 1024 pixels, one image, 26 steps and CFG 3.5, then run to receive the generated image batch. Width and height use pixels in multiples of 16; their ratio determines the image aspect ratio. Increasing dimensions or image count increases memory requirements. The selected output returns images through Sweet Tea's preview transport rather than a SaveImage file-saving node. Advanced controls include seed, prompt guidance, steps, sampler, scheduler, flow shift and noise schedule fraction. Seed -1 randomizes each run; use a fixed seed when comparing settings. CFG controls the strength of positive and negative prompt guidance. The sampler chooses the sampling algorithm, and the scheduler distributes noise levels across the steps. Flow shift adjusts the flow model's sampling timestep shift. Euler, beta scheduling and flow shift 1 are the supplied starting settings. Keep noise schedule fraction at 1 for the full generation schedule. Lower values shorten that schedule. This workflow starts from an empty latent and has no uploaded-image or image-to-image mode. Its fixed model setup uses Chroma1-HD-fp8mixed.safetensors, t5xxl_fp8_e4m3fn_scaled.safetensors and ae.safetensors.
NetaYume Lumina Text to Image
Generate anime-style images from text using the fixed NetaYume Lumina model. Enter a positive prompt and adjust the negative prompt, image dimensions, and batch size. Advanced settings provide seed, steps, guidance, sampler, scheduler, denoise, and model shift controls. Fixed model-specific prefixes are added to both prompts.
Coloring Book Style
Trained on a variety of coloring book images, it produces clear line drawings with plenty of white space for your crayons. Adding the prompts white background, b&w and the negatives: color, gradient, grey seem to...
Pure Eros Face
This embedding will generate good-looking girl faces , close concept of kpop idols or Instagram girls . For SD1.5 based models.
djz CYBER KAN V0
Curated Collection of Manga Illustrations from my videogame series "BIG KAN", part of the Paradise Universe.
Handsome male models' face (18-30) /realism/
if you want more stronger changes and push the images towards more 18-20 years face, use it around 1.5-2.00
Restyle an image with Seedream 5.0 Lite
Restyle a source image using a second image as a visual reference and a written instruction. Upload both required images: the picture to change as Source image and an example of the desired look as Style reference. In Edit instruction, refer to them as the first and second images respectively. The default instruction requests style transfer; edit it to describe the changes you want and the details to preserve. For a first run, keep the square 2048 × 2048 preset and Maximum images at 1, then run the tool to view the generated image. Select Custom under Output size to activate Width and Height, measured in pixels. Width accepts even values from 1024 to 6240; Height accepts even values from 1024 to 4992. Maximum images at 1 produces one result. Higher values request between one and the chosen maximum number of related images; describe the desired variations in the instruction. With these two reference inputs, use at most 13 generated images because input and generated images together cannot exceed 15. Enable Require complete results to abort on missing or errored requested images; the default accepts available results. Seed -1 randomizes each run; enter a fixed seed to control the generation seed. AI watermark asks the provider to add an AI-generated watermark. Prompt reasoning stays enabled because it cannot be disabled when reference images are supplied. This tool uses the Seedream provider API, sends the reference images and instruction to the provider, and requires configured provider access with provider billing. Generated images are returned through the preview output. Uploaded alpha masks are not connected, so transparency preservation is not supported.
negative_hand Negative Embedding
Currently, there are more and more negative embedding, while many are also very good and easy to use. However, almost all of them currently have a big problem... They change the main or initial artstyle of the used...
Two-speaker dialogue
Enter each speaker's dialogue and choose a voice for each entry. ElevenLabs generates the entries together, and the result is saved as MP3. Stability controls emotional variation versus consistency. Leave the language code empty for automatic detection. Voice cloning is not connected to this output.
Winter Style
This is a DreamArtist Textual Inversion Style Embedding trained on images of a snowy winter forest/cabin scene. You can get amazing winter colors and snow effects in your images by calling Style-Winter in the...
Mage-Flow Turbo image edit
Edit one uploaded image with written instructions using the fixed Mage-Flow Edit Turbo model. Describe what to keep and what to change, such as content, background, lighting or style; no painted mask is used. For a first run, upload your source image, replace the example instruction with an edit suited to it, and keep Image count at 1. Leave the source sampling configuration at four steps, guidance scale (CFG) 1, Euler, the simple scheduler and denoise fraction 1. Seed -1 chooses new sampling noise each run; use a fixed seed when comparing instructions or settings. Resize source image is enabled by default at approximately one megapixel. Target megapixels controls the processed reference area, retaining aspect ratio and aligning dimensions to multiples of 16. Reference resampling selects the interpolation used for this resizing. Turn off Resize source image to condition on the original reference dimensions. Match source size makes the output follow the processed reference dimensions, including any reference resizing. Turn it off to use Width and Height instead. Enter manual dimensions in pixels divisible by 16; the supplied guidance recommends 512–2048 pixels per side. Very wide or tall references can exceed that range under automatic sizing, so choose manual dimensions when needed. Image count generates a batch of edited images; larger batches require more memory. Negative prompt describes attributes to discourage and has little or no influence at CFG 1. Increasing guidance makes negative conditioning relevant. Advanced sampling controls remain available for comparisons, but more steps or stronger guidance do not guarantee a better Turbo edit. Denoise fraction changes the sampling schedule; it is not established here as a reliable source-image preservation strength. This graph accepts one reference image and returns the decoded edited image batch. It does not expose multiple references or a selection mask. Uploaded transparency is not preserved because only the loader's RGB image output is connected.
Reference-guided relighting
Generate one relit image from a subject image and a lighting reference, or from the subject and text instructions alone. For a first run, upload the subject as image 1 and a lighting reference as image 2. Keep the lighting-reference stage enabled, retain the supplied prompt, and leave Lightning mode off. The supplied Chinese prompt asks to remove image 1's original lighting and relight it using image 2's lighting and color tone. Standard sampling uses 20 steps and CFG 4. Enable Lightning mode to select the additional 4-step Lightning LoRA and its separate settings, initially 4 steps and CFG 1. Switching modes selects the current values for that mode; it does not reset edited values. Each mode's steps and CFG remain independently adjustable. Lightning mode uses the 4-step model artifact, even if an upstream source note mentions an 8-step artifact. The light-migration LoRA remains active in both modes. To relight without a reference, disable the lighting-reference stage and rewrite the instructions to describe the desired lighting directly, without referring to image 2. No reference upload is needed in this mode. A subject upload is always required; when the reference stage is enabled, upload a reference too. Advanced controls include an optional negative prompt, both LoRA strengths, denoise, seed, sampler, scheduler, guidance normalization and sampling shift. Lower denoise retains more of the initial subject's structure and can limit the lighting change. Use seed -1 for a fresh random seed, or a fixed seed to compare edits with the same initial noise. LoRA strengths control their model contributions; Lightning strength only affects Lightning mode. Changing an inactive mode's steps or CFG affects subsequent runs in that mode. The subject is automatically scaled before conditioning and sampling, with no independent output-size control. The Pipe returns one generated image. Relighting may change subject details, and uploaded transparency is not preserved because only the image loaders' RGB outputs are connected.
Grok image edit
Create one or more Grok-edited images from an uploaded source image and text instructions. For a first run, upload the image you want to edit, replace the robot example prompt with your own instructions, and use the default grok-imagine-image model, 1K resolution, and one result. Describe both the changes you want and any details or framing you want retained, then run the tool. Choose among grok-imagine-image, grok-imagine-image-quality, and grok-imagine-image-pro. Resolution offers the provider's 1K or 2K setting rather than an exact width and height. Request 1–10 edited results from the same source image and instructions; this does not add source images. Aspect ratio stays automatic because the provider's explicit aspect-ratio setting requires multiple source images, while this tool supplies one. Seed controls whether the provider node re-runs; results remain nondeterministic even when a seed is repeated. Leave it at -1 to randomize the re-run seed each run. Uploaded transparency is not preserved: only the loader's RGB image output reaches the editor.
Magnific portrait skin enhancement
Produce a skin-enhanced portrait using Magnific's creative mode. Upload a portrait, leave sharpening at 0 and smart grain at 15 for your first run, then run the Pipe to receive the enhanced image. Sharpening controls sharpening intensity; smart grain controls grain intensity. Both use a 0–100 scale and can be adjusted independently. Creative mode allows artistic changes to the portrait's appearance. This Pipe supports creative mode only; faithful and flexible modes are unavailable. Uploaded transparency is not preserved because processing uses the loader's RGB image output. Processing sends the portrait to the Magnific partner API, requires provider authentication configured in the runtime, and uses provider-priced processing.
SD3.5 Large blur-guided image generation
Generate new images from a prompt and an already preprocessed blur reference using SD3.5 Large and its blur ControlNet. Upload the required reference and describe the desired subject, appearance, materials, and details in the positive prompt. This tool does not blur or otherwise preprocess the upload. For a first run, replace the example prompt with your description, upload a reference with the intended composition, and choose output dimensions matching its aspect ratio. Width and height are in pixels and use multiples of 16. Start with one image and the defaults: 1024 × 1024, control strength 1, 30 steps, CFG 4, Euler sampler, Simple scheduler, and denoise fraction 1. Control strength determines the influence of the reference. Lower it for more freedom; set it to 0 for generation without reference guidance, while still supplying the required upload. Control start and end select the portion of sampling during which guidance applies: 0 is the beginning and 1 is the end. Keep start below end. Open Sampling for variation and tuning. Seed -1 selects a random seed each run; use a fixed seed when comparing settings with the same initial noise. CFG controls prompt guidance. Steps, sampler, scheduler, and denoise fraction change the sampling process. Keep denoise at 1 for the initial run. Image count generates a batch using the same prompt, reference, and dimensions; larger batches and dimensions require more memory. The checkpoint and blur ControlNet are fixed in this tool. Generation starts from an empty latent rather than an encoded copy of the upload, so this is not pixel-preserving restoration. Lowering denoise does not turn it into image-to-image restoration. There is no editable negative prompt: its text conditioning is zeroed internally before ControlNet is applied. Uploaded alpha is not used.
Seedance 2.0 text to video
Generate one video from a scene description, with optional generated audio. No uploaded media is required. For a first run, use the supplied five-second fantasy prompt with Seedance 2.0, 720p, 16:9 and Generate audio enabled. Alternatively, replace the prompt with your subject, setting, action, visual style and camera movement. Include desired sound when audio is enabled. This workflow calls the ByteDance provider API, requires runtime authentication and uses provider-priced generation. Choose Seedance 2.0, Seedance 2.0 Fast or Seedance 2.0 Mini. Resolution is a text field: enter a value supported by the selected model rather than assuming every resolution works with every model. Aspect ratio sets the frame shape; adaptive leaves that choice to the provider. Duration accepts 4–15 seconds. When changing it, update any duration or shot timestamps in the prompt as well. Prompt wording such as '8K' does not replace the Resolution control. Disable Generate audio for a silent result. Watermark requests a provider watermark. Seed defaults to -1 for randomization on each run. A fixed seed does not reproduce the same video; the source describes generation as non-deterministic regardless of seed. The generated video is returned as one preview result. Under advanced Video encoding, leave File type and Codec at auto for automatic selection, or choose a supported container and compatible codec. This output uses Sweet Tea's preview transport rather than the original SaveVideo stage.
GPT Image edit and generation
Edit a reference image or generate images from text using GPT Image. For a first run, upload one 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: GPT Image 1.5, Auto output size, Auto background, High quality and one image. For text-only generation, disable the Source image stage and replace the editing instruction with a complete scene description. An upload is required while that stage is enabled. Model selects GPT Image 1, GPT Image 1.5 or GPT Image 2. Output size requests pixel dimensions or Auto. GPT Image 2 enables additional sizes and custom dimensions. To use custom dimensions, select GPT Image 2 and Custom output size, then enter width and height from 1024 to 3840 pixels in multiples of 16. These fields have no effect with other output-size selections. Background requests automatic, opaque or transparent background handling. Quality selects Low, Medium or High and affects generation time and cost. Image count requests 1 to 8 images using the same instruction and optional reference; the selected output returns those images. This tool connects one RGB reference image. Uploaded alpha transparency is not preserved, and the loader's separate alpha-mask output is unused. Although the generation node supports an optional inpainting mask, this tool does not connect or expose one, so it has no mask-based editing mode. The source seed setting is hidden because the backend does not implement it; it cannot make results reproducible.
fal.ai LTX Video
Describe a scene and its motion to generate a video through fal.ai's LTX Video 13B Distilled endpoint. This workflow accepts text only and returns the generated video. For a first run, enter a prompt describing the subject, setting, action, and camera movement. Keep the source defaults of 480p, 16:9, and 121 frames. The negative prompt provides a starting list of visual and motion defects to discourage. Resolution and aspect ratio set the video shape. Frame count sets its length in frames; this workflow does not expose FPS, so frame count is not a duration in seconds. Frame count must be entered as an integer, without units. Alternative shape values and frame counts must be supported by the provider. Seed defaults to randomization on each run. A fixed seed is available for exploring variations under otherwise similar settings, without a guarantee of identical provider results. Generation requires configured fal.ai access.
