ComfyUI Node
Hunyuan 3 Generate (Low VRAM Budget)
A ComfyUI node in HunyuanImage3 with 13 inputs and 4 outputs.
Hunyuan 3 Generate (Low VRAM Budget)
- model
- image
- rewritten_prompt
- status
- trigger
◄promptA beautiful landscape►
◄seed0►
◄steps40►
◄resolution▾►
◄guidance_scale7.5►
◄offload_modesmart►
◄post_actionsoft_unload_to_cpu►
◄enable_prompt_rewritefalse►
◄rewrite_stylenone►
◄api_urlhttps://api.deepseek.com/v1/chat/completions►
◄model_namedeepseek-chat►
◄gpu_budget_gb-1.0►
CategoryHunyuanImage3
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| model | HUNYUAN_MODEL | — | |
| prompt | STRING | A beautiful landscape | — |
| seed | INT | 00–18446744073709550000 | — |
| steps | INT | 401–100 | Number of sampling steps. 30-40 is recommended. |
| resolution | COMBO | 20 options: Auto (model default), 720x1280 - Portrait 720p (0.9MP), 864x1152 - Portrait 3:4 (1.0MP), 1024x1024 - Square 1K (1.0MP), 1152x864 - Landscape 4:3 (1.0MP), 1280x720 - Landscape 720p (0.9MP), +14 | |
| guidance_scale | FLOAT | 7.51–20 | Classifier-free guidance scale. Higher = closer to prompt, Lower = more creative. |
| offload_mode | COMBO | smart | For NF4/INT8 models, offloading is handled automatically by the loader (device_map). This setting is ignored for quantized models. |
| post_action | COMBO | soft_unload_to_cpu | After generation: soft_unload_to_cpu recommended for Low VRAM to free GPU for downstream tasks |
| enable_prompt_rewriteopt | BOOLEAN | false | — |
| rewrite_styleopt | COMBO | none | 3 options: none, en_recaption, en_think_recaption |
| api_urlopt | STRING | https://api.deepseek.com/v1/chat/completions | — |
| model_nameopt | STRING | deepseek-chat | — |
| gpu_budget_gbopt | FLOAT | -1.0-1–128 | Override loader GPU budget for this run (GB). -1 uses loader default. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| image | IMAGE | — |
| rewritten_prompt | STRING | — |
| status | STRING | — |
| trigger | * | — |