FreedomGPT Image Understand

FreedomGPT Image Understand is the "describe what you see" node: you feed it an image and a question, and it returns a text answer from FreedomGPT's vision model. "Is this person wearing a hat?" "Transcribe the text in this logo." "What's wrong with this render?" It's the read-the-image counterpart to the pack's cloud image generation node, and it opens up the whole auto-prompt / image-analysis workflow family - a vision model describing an output, that description feeding a prompt builder, a new image generated from it. Round-trip loops without any local captioning model.

The mechanism is standard vision-API plumbing: your ComfyUI IMAGE tensors (up to five, via image plus image1–image4) get converted to PIL and base64, sent to https://chat.freedomgpt.com/api/v1 with the prompt, and the model's text answer comes back. Because it's an API call, there's no local model download and no VRAM - but there's also a round-trip, an API key, and, for anyone outside regions with easy access, possibly a proxy to think about.

Inputs that matter

• api_key - or the shared key file / FREEDOMGPT_API_KEY env var.
• prompt - the question or instruction about the image. system_prompt shapes the overall tone.
• model_name - defaults to liberty; verify against FreedomGPT's current vision model list if it errors.
• temperature (0–2), top_p, top_k, max_tokens - the usual sampling knobs. Defaults are sensible; most people never touch them.
• chat_template - the message formatting template, with {{system_instruction}} and {{prompt}} placeholders. The default works; changing it is for advanced users.
• seed, proxy_url, timeout - reproducibility, proxy support, and request timeout (0 = none).

Outputs

• text - the model's answer, a STRING you can pipe into a prompt builder, a Switch, or StringConcat.
• history - the conversation history, so you can chain follow-up questions across nodes.
• payload - the raw request payload, *-typed, useful for debugging what was actually sent.

Install & key setup

Part of ComfyUI-YogurtNodes:

cd ComfyUI/custom_nodes
git clone https://github.com/yogurt7771/ComfyUI-YogurtNodes.git
cd ComfyUI-YogurtNodes
pip install -r requirements.txt

Restart ComfyUI; the node lives under "Yogurt Nodes" → LLM. Keys resolve in this order: the node's api_key field → api_key.json in custom_nodes/ComfyUI-YogurtNodes/yogurt_nodes/llm/ ({"freedomgpt": "YOUR_KEY"}) → the FREEDOMGPT_API_KEY env var.

Common issues

• "No text comes back" - check the text output is actually empty vs. just short, then check payload to see what got sent.
• "It can't see my image" - make sure an image is actually wired into one of the image ports; the node only sends what's connected.
• "Auth / timeout errors" - same key-resolution and proxy story as the generation node: check key sources in priority order and set proxy_url if the API is blocked in your region.
• "The answer is generic" - vision models are literal; ask precise questions. "What color is the subject's shirt?" beats "describe this."

Honest take: if you're already paying for a big API, you may have a preferred vision provider - the pack offers Gemini, OpenAI, Grok, OpenRouter, and Vertex variants too. This node is the right call specifically when FreedomGPT is your provider of choice or your only reachable one.