- READMEs (en/ja/zh): the chat is with an LLM / VLM; the optional image judge is
a decision model (not called VLA). Mention the UI row still labeled "VLA".
- Code comments: use "decision model" instead of VLA (no functional change).
- docs/thumb.png: updated artwork ("Decision" instead of "VLA"), resized to
800x400 (Registry icon limit).
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The routes no longer accept a URL from the browser. The server reads the Ollama
address from the LIVE_CHAT_STREAM_OLLAMA_URL environment variable or from
<user>/live_chat_stream/config.json ("ollama_url"), defaulting to
http://127.0.0.1:11434. The host allow-list / DNS checks are no longer needed
and are removed. The UI shows that the address is a server setting.
Addresses the review comment on Comfy-Org/ComfyUI-Manager#3340.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
- Resolve host names and require every address to be loopback / private
- Reject link-local, multicast, unspecified and public IPv6 literals
- Reject URL credentials; do not follow redirects
- Document the --listen exposure in the READMEs
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Chat with an LLM / VLM / VLA (Ollama) and turn the streaming reply into images.
- Two prompt lanes: A (<prompt> block) and B (thinking / reply / <mind>)
- Character presets stored on the server, system prompt modal
- Image input / output (I2I), VLM attachments, VLA judge
- VRAM management: manual unload and auto-free before generation
- i18n: English / Japanese / Simplified Chinese
- Apache-2.0
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
