frameforgeYOUR CREATIVE STUDIOBRING YOUR OWN ENGINE
Get ready to create.
Already have ComfyUI? Connect it here, choose a workflow and get just its missing files. Prefer an online engine? Start with Comfy Cloud or Civitai instead.
START WITH THE SETUP YOU HAVE
Choose how you want to create
Online generation needs an account with the provider, without a local GPU or model download.
I already have ComfyUI
Use your existing engine and model folders.
Connect my installation ↗02 / INSTALLI want to run locally
Follow the install guide, then get only the files your recipe needs.
Help me set it up ↗03 / CLOUDComfy Cloud only
Prepare a real graph and run it in your browser.
Create without local setup ↗04 / CIVITAIUse Civitai’s generator
Take your prompt and model references to Civitai.
Prepare a Civitai image ↗01 / CONNECT
Connect your ComfyUI
For local generation, install ComfyUI and start it first, or reuse your existing installation. If you prefer online generation, choose a Cloud or Civitai route above and skip this section.
Download ComfyUI ↗ · Installation guide ↗
Checking your setup…
Using a model folder on another drive?
Keep your models where you have space. Add that folder through ComfyUI’s extra model paths configuration, then restart ComfyUI. Frameforge sees the files that your connected engine advertises.
ComfyUI model folder guide ↗Prefer online generation?
No local models needed. Choose a matching paid workflow, download its JSON and inputs, then import them on Comfy Cloud. Sign in and pay there; this studio stores no Cloud key.
02 / CHECK YOUR RECIPE
Only get what you need
ComfyUI says models are missing?
Import the workflow JSON into ComfyUI. On the setup-error notice, choose View details, then Issues → Missing Models. Review the listed filenames and folders. Use Download for one file or Download all for the recipe; large models may take several gigabytes.
These controls were checked with the Z-Image graph in a separate ComfyUI installation with empty model folders. If your edition downloads through the browser, save each file in its listed folder. Refresh the model list or restart ComfyUI, then check your recipe here again.
YOUR WORKSPACE
Navigation & execution
Choose your desktop navigation. Narrow windows use a labelled Menu button automatically.
Open weights only limits execution to the reviewed local graphs and processors. Partner routes stay hidden until you turn it off. A locked launcher keeps this setting on.
Processor software, separate from weights
Check just the CPU tools this workflow needs. These commands run from your Frameforge folder and use isolated environments. They do not download models or alter your ComfyUI Python.
node scripts/setup-processor.mjs media --check
node scripts/setup-processor.mjs vision --check
To install a missing profile, replace --check with --install. Review the pinned requirements first. GPU speech, lip sync and advanced model processors still use their separate source guides; this is not an installer for all processors.
NATIVE COMFYUI
Your full Studio, inside ComfyUI
Open Images, Video, Audio, Library and Guide without leaving your graph. Install the extension below, start both apps, then choose the Frameforge sparkle icon in ComfyUI → Quick tools. Use Studio, Split or Graph to switch views without replacing your workflow. The compact Quick tools menu jumps to video, image and audio editors. Open graph in ComfyUI beside a selected workflow copies prepared inputs into a new graph tab; the regular website offers a one-click workflow package instead. Connection lets you choose another local Studio address.
Use the matching Frameforge output node for IMAGE, VIDEO or AUDIO. Run it, then choose Send saved media to Frameforge to open the embedded editor. PNGs open in Images, MP4s in Video and WAVs in Audio. Imports also appear in My work.
Unzip into your ComfyUI custom_nodes folder, then restart ComfyUI when idle. Both apps run on this computer. No new weights are needed. Video uses ComfyUI’s VIDEO type; connect Create Video first when your graph produces image frames. Audio exports one PCM WAV item.
03 / BRING YOUR OWN AI
Use the AI you already have
Ollama is optional. Power Guide with LM Studio, Ollama, llama.cpp, vLLM or another local compatible server. Or keep your assistant in Antigravity, Hermes, Claude or another MCP client and give it Frameforge tools. Guided matching works without any language model.
After connecting a model server, open Guide → Answer with and choose a chat model. Your model, its license, GPU offload and unloading stay under that server’s control. The website guide suggests verified tool links; the separate MCP connection can enable local creation.
If your local server requires authentication, set FRAMEFORGE_ASSISTANT_API_KEY in the studio environment before startup. No key is requested or saved by this page. This connector accepts only local HTTP server addresses and does not silently connect to a hosted API.
The Open weights only profile keeps this website guide in catalog-matching mode until a particular chat-model dependency is audited. A local URL alone cannot certify open weights or that a server never proxies remotely.
SHARE THE LAYER, KEEP YOUR FILES
Made for a fresh install
The GitHub source package includes the interface, workflow JSON and model download manifests. Your models, Python environments, generated media, personal settings and API keys stay outside it. Start the website with Node.js 22 or later; connect ComfyUI when you are ready.
Standalone speech, masking and video finishing also need their documented local Python tools. A connected ComfyUI alone does not install those processors.
Explore the workflow library ↗