ComfyUI Node
Ollama Chat (Local)
Run a local Ollama model with a text prompt and optional image batch/list. Connect Response to any STRING-compatible node. Use Edit > Ollama: Refresh Models after downloading a new model.
Ollama Chat (Local)
- images
- Response
◄ollama_urlhttp://localhost:11434►
◄model▾►
◄prompt►
◄seed0►
◄temperature0.70►
◄max_tokens512►
◄context_size4096►
◄system_prompt►
◄thinking▾►
◄keep_alive▾►
◄image_format▾►
◄image_max_side1024►
◄jpeg_quality90►
◄max_images8►
◄timeout_seconds300►
◄cache_mode▾►
CategoryVicNodes/Ollama Bridge
Inputs (17)
| Name | Type | Default | Description |
|---|---|---|---|
| ollama_url | STRING | http://localhost:11434 | Address of the Ollama service as seen by the ComfyUI backend. |
| model | COMBO | Locally downloaded Ollama model. Refresh via ComfyUI Edit > Ollama: Refresh Models. | |
| prompt | STRING | Message sent directly to the model. Also accepts images connected below. | |
| seed | INT | 00–2147483647 | Ollama sampling seed. ComfyUI provides fixed/increment/decrement/randomize. |
| temperature | FLOAT | 0.700–2 | Lower is more focused, higher is more varied. |
| max_tokens | INT | 5121–16384 | Maximum generated tokens (Ollama num_predict). |
| context_size | INT | 4096512–131072 | Context window (Ollama num_ctx). Larger sizes use more memory. |
| system_prompt | STRING | Optional system instruction; leave blank to use the model's normal behavior. | |
| thinking | COMBO | Override thinking for models that support Ollama's think option. | |
| keep_alive | COMBO | How long Ollama keeps the model in memory. 0 unloads; -1 keeps indefinitely. | |
| image_format | COMBO | JPEG sends less data; PNG preserves exact text and fine details. | |
| image_max_side | INT | 1024256–4096 | Resize large reference images before sending them to Ollama. |
| jpeg_quality | INT | 9050–100 | JPEG quality only; ignored in PNG mode. |
| max_images | INT | 81–32 | Safety limit across all images, including batches and lists. |
| timeout_seconds | INT | 30010–3600 | Maximum wait time for Ollama to generate its answer. |
| cache_mode | COMBO | Use ComfyUI caching for speed, or force an Ollama call on every queue. | |
| imagesopt | IMAGE | Optional Load Image, IMAGE batch, or IMAGE list. Requires a vision model. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| Response | STRING | — |