GrassVison
moduqishi/GrassVison · model-adapters
Filling a pure text large model with an original vision: flow of real thinking ..cross the rotational invisibility ..indicative evidence compatible with SVG graphics ..openAI/Anthropic/Resones
★ 3GitHub stars
0Forks
2026-08-14Last updated
PythonLanguage
—License
Key features
- Monitoring the environmentSourceSeek V4 Flash . Image model MiniMax-M3 (minimax channel) Visual channels vivo / cpa failure transfer.
- The visual model's analytical reasoning, the source model's thought chain, the final answer, a single line of SSE flow, seamless connection.
- Flowing chain of real thinkingvisual reasoning, ⁇ source model thinking, ⁇ re-thinking 2 ⁇ answer, seamless flow, the frame is what's really happening.
- Positioning-Zooming-Rereadingimage. grounding zoom: Coordinate frame + Local cropping to magnify secondary reading
- When used in DeepSeek Harness, you need to manually open enhanced model image support with the configuration file.
Install command
dsh plugin --profile web add github:moduqishi/GrassVison