ComfyUI Node Runs on cloud
Molmo Vision-Language Model
A ComfyUI node in VLM Nodes/Molmo with 9 inputs and 1 output.
Molmo Vision-Language Model
- image
- STRING
◄promptDescribe this image in detail.►
◄model_name▾►
◄memory_mode4-bit Quantized (15GB+ Required)►
◄max_new_tokens200►
◄temperature0.7►
◄top_p0.9►
◄top_k50►
◄use_autocasttrue►
CategoryVLM Nodes/Molmo
Inputs (9)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | Input image to be analyzed by Molmo | |
| prompt | STRING | Describe this image in detail. | Instructions for the model. Be specific about what aspects of the image you want analyzed. |
| model_name | COMBO | ⚠️ WARNING: These are very large models requiring significant RAM/VRAM. Start with MolmoE-1B. | |
| memory_mode | COMBO | 4-bit Quantized (15GB+ Required) | Controls RAM/VRAM usage. Use most aggressive option that works on your system. |
| max_new_tokens | INT | 2001–2048 | Maximum tokens to generate. Higher values need more VRAM. Start small (200) and increase if needed. |
| temperature | FLOAT | 0.70.1–2 | Controls randomness. Lower (0.1-0.5) = more focused, higher (0.8-2.0) = more creative. |
| top_p | FLOAT | 0.90.1–1 | Nucleus sampling. Lower = more focused on likely tokens, higher = more diverse vocabulary. |
| top_k | INT | 501–100 | Limits token choices to top K most likely. Lower = more focused, higher = more variety. |
| use_autocast | BOOLEAN | true | Enables mixed precision. Keeps quality while reducing VRAM usage. Recommended ON. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| STRING | STRING | — |