This is a fast image editing application based on FLUX.2 Distilled 8-step + ReferenceLatent + FluxGuidance + Image Reference Editing. Users upload an original image and input specific editing instructions. The system performs targeted replacement, enhancement, or commercial upgrades while preserving the original composition, perspective, subject relationships, and basic photographic texture.
The biggest difference between this and regular text-to-image is that: instead of generating a new image from scratch, it edits based on the input image.
For example, in the current default case, the user uploads a regular pasta image, and the prompt asks to upgrade it to "Michelin-grade lobster and truffle pasta". The workflow strives to retain the original bowl angle, table composition, shallow depth of field, and food photography basics, while replacing or enhancing ingredient details such as lobster chunks, black truffle slices, cream sauce gloss, parmesan foam, fresh basil, steam, and high-end restaurant plating.
There are three core advantages to this workflow. First, fast speed: it uses the FLUX.2 8-step distilled model, making it ideal for quickly testing editing directions without traditional FLUX long-step sampling. Second, more stable composition retention: ReferenceLatent makes the model reference the structure of the original image rather than completely recomposing it. Third, clear commercial editing direction, making it especially suitable for product image upgrades, food image retouching, e-commerce primary image transformation, advertising material style changing, packaging visual testing, and rapid social media cover image iteration.
When using it, it is recommended to focus on controlling four items: the input image determines the base composition, the editing prompt determines what to change, Guidance determines execution strength, and Steps determine the balance between speed and detail.
If it is just a minor modification, the prompt should emphasize "preserve original composition"; if you want a strong content replacement, you can make main_edit more explicit. The current negative prompt already includes restrictions such as low quality, blurriness, messy food, plastic look, over-sharpening, text, and watermarks, making it suitable for commercial image editing scenarios.
When publishing as a RunningHub application, it is recommended to keep the frontend very simple: Upload Image + Editing Requirements + Prompt Adherence Strength + Seed + Output Image. Models, VAE, CLIP, ReferenceLatent, and samplers do not need to be shown to regular users.
🎁Claim RH Coins first, then experience the workflow! Top right avatar → Invitation Code → Enter [rh-v1111] to instantly receive 1000 RH Coins. Log in daily to get another 100 RH Coins~ For more ComfyUI workflows, tutorials, and gameplay, welcome to follow the official WeChat account (AIKSK), with simultaneous updates on Douyin / Bilibili / Xiaohongshu / YouTube (AI-KSK) ✨


No creations yet

No creations available.