FrameLoom — Krea2 | Cloud & Local LLMs Inside ComfyUI
FrameLoom brings an LLM chat and Krea 2 image generation together in one workspace. Discuss an idea, build a character, refine the composition, and develop your prompt step by step — then send it to generation when you’re ready.
Compact control panels, a shared LoRA library, and a large image preview keep the workflow easy to manage without constantly panning across the graph.
🔞 Support for NSWF content depends on your chosen LLM, image model, and provider policies.
⚠️ Content Disclaimer
FrameLoom is a creative tool. You are responsible for your prompts, reference materials, generated content, and how that content is used, shared, or published. The author does not review or endorse user-generated content and accepts no responsibility for misuse of this workflow.
Please follow applicable laws, respect third-party rights, and comply with the licenses and terms of the models and API providers you use.
✨ Key Features
Chat inside your workflow. Talk to the LLM independently of image generation. Explore ideas and refine details before creating an image.
Cloud and local models. Connect to OpenRouter, LM Studio, Ollama, or another server with an OpenAI-compatible API.
Automatic local model selection. If your server lists one model, it is selected automatically. If several are available, you can choose from a list.
Generation-ready English prompts. The node extracts the complete prompt from the LLM’s response and places it in a dedicated generation field.
Editable prompts. Make your own adjustments before generating, or restore the latest LLM-written version. Manual edits apply to image generation only.
Iterative editing. Built-in instructions tell the LLM to preserve details you haven’t asked to change and clarify conflicting requirements.
Custom system prompts. Save named presets and switch between them in Settings. The core interaction rules and response format are built into the node.
Adjustable timeouts. Set separate limits for connecting, waiting for incoming data, and completing the entire request.
🧠 Three Assistant Modes
Direct prompt. Describe your scene and refine the result through normal conversation.
Clarify the request. The model can ask questions with contextual answer choices. Pick an option, write your own answer, or let the model decide.
Character builder. Develop a character’s appearance, clothing, and other details through successive questions and an editable character card. The final prompt is created after you confirm the card.
🖼 Images in Chat
Attach multiple images using the + button. Choose files from your computer or select thumbnails from the latest generation results.
Discuss a generated image and use visual examples to explain what you want to change. Image analysis requires a vision-capable LLM. Attachments are kept only for the current tab session.
📚 Shared LoRA Library
Browse your local LoRAs with search, folder filters, and a selected-only view.
Save descriptions and exact trigger words once and reuse them across workflows.
Pass information about active LoRAs to the LLM when preparing a prompt.
Adjust each selected LoRA independently: the default strength is 1.0, the + / − buttons change it in 0.1 steps, and you can enter a value manually.
Add new files to
models/lorasand refresh the catalog.
📐 Canvas Size & Current Settings
Choose an aspect ratio preset, enter custom dimensions, or swap width and height. The chat and image generator use the same canvas dimensions.
The LLM receives the current aspect ratio without being instructed to force a full-body shot or landscape scene. After changing the canvas, selected LoRAs, or their strengths, use Refresh for current settings to prepare an updated prompt.
⚙️ Generation & Finishing Tools
Universal Loader: select the diffusion model, text encoder, and VAE in one panel, with search and file-format filters.
GGUF support for compatible diffusion models and text encoders through ComfyUI-GGUF.
Compact generation controls: sampler settings, steps, CFG, denoise, batch size, and an optional second sampling pass.
Seed controls directly in the chat node: fixed, random, increment, or decrement.
FaceDetailer: optional face refinement with its own settings.
Neural 2× upscaling: a separate stage for increasing image resolution.
Three independent switches below the chat: Upscale, Second pass, and FaceDetailer. All three are disabled by default in the public example.
💾 Preview & Saving
A large image preview sits beside the chat. Results are saved to ComfyUI’s output folder, with additional options to download a PNG or automatically save a copy to a folder chosen through your browser.
Folder selection requires a compatible browser, such as Chrome or Edge, and permission to write to the selected folder.
🌐 Five Interface Languages
Switch between English, Russian, Chinese, Japanese, and Korean using a dedicated language node. This changes the node pack’s interface while leaving chat and prompt content unchanged.
💡 My Recommendation
I recommend using larger models through OpenRouter, especially for complex scenes, iterative editing, and the character builder. These modes benefit from a model that can follow detailed instructions and keep track of many requirements.
You can also connect a local LLM. The experience will depend on the model you choose and the memory available on your system.
You can use OpenRouter for free by creating an account and generating an API key. Then, either select a model from the list of free options or enter 'openrouter/free' into the 'model' field.
📦 Installation
The workflow JSON and custom node archive are provided separately. The archive includes a README and separate English installation guides for Windows and Linux. Model weights are not included. GGUF loading, FaceDetailer, and neural upscaling require their respective additional components.
The custom node code is released under the MIT license. This version has been tested on Linux; Windows runtime compatibility has not yet been independently verified.