ComfyUI Node
ArchAi3D_Qwen_Encoder_V3
A ComfyUI node in ArchAi3d/Qwen/Encoders with 21 inputs and 4 outputs.
ArchAi3D_Qwen_Encoder_V3
- clip
- vae
- image1_vl
- image2_vl
- image3_vl
- image1_latent
- image2_latent
- image3_latent
- conditioning
- latent
- formatted_prompt
- recommended_cfg
◄prompt—►
◄system_prompt►
◄conditioning_balance▾►
◄conditioning_balance_override►
◄manual_context_strength1.00►
◄manual_user_strength1.00►
◄image1_labelImage 1►
◄image2_labelImage 2►
◄image3_labelImage 3►
◄image1_latent_strength1.00►
◄image2_latent_strength1.00►
◄image3_latent_strength1.00►
◄debug_modefalse►
CategoryArchAi3d/Qwen/Encoders
Inputs (21)
| Name | Type | Default | Description |
|---|---|---|---|
| clip | CLIP | Qwen-VL CLIP model for encoding text and vision tokens | |
| prompt | STRING | Text prompt (vision tokens inserted automatically in ChatML format) | |
| system_prompt | STRING | Optional system prompt (wrapped in ChatML <|im_start|>system block) | |
| conditioning_balance | COMBO | V3 PRESET: Choose conditioning balance (Image-Dominant → Text-Dominant). Works great with ConditioningAverage! | |
| conditioning_balance_override | STRING | OPTIONAL: Connect ⚖️ Conditioning Balance node here to override the preset above. Leave empty to use the dropdown. | |
| manual_context_strength | FLOAT | 1.000–3 | CUSTOM ONLY: Manual context strength (only used when preset = Custom). Extended range: 0.0-3.0 for extreme conditioning control |
| manual_user_strength | FLOAT | 1.000–3 | CUSTOM ONLY: Manual user strength (only used when preset = Custom). Extended range: 0.0-3.0 for extreme conditioning control |
| image1_label | STRING | Image 1 | Custom label for Image 1 |
| image2_label | STRING | Image 2 | Custom label for Image 2 |
| image3_label | STRING | Image 3 | Custom label for Image 3 |
| image1_latent_strength | FLOAT | 1.000–2 | Image1 latent strength (1.0=normal, <1.0=weaker, >1.0=stronger) |
| image2_latent_strength | FLOAT | 1.000–2 | Image2 latent strength (1.0=normal, <1.0=weaker, >1.0=stronger) |
| image3_latent_strength | FLOAT | 1.000–2 | Image3 latent strength (1.0=normal, <1.0=weaker, >1.0=stronger) |
| debug_mode | BOOLEAN | false | Enable console logging (shows preset values, strengths, shapes) |
| vaeopt | VAE | VAE for encoding reference latents (required if using latent images) | |
| image1_vlopt | IMAGE | Image 1 for vision encoder (RGB only, expects correct size) | |
| image2_vlopt | IMAGE | Image 2 for vision encoder (RGB only, expects correct size) | |
| image3_vlopt | IMAGE | Image 3 for vision encoder (RGB only, expects correct size) | |
| image1_latentopt | IMAGE | Image 1 for reference latent (RGB only, expects correct size) | |
| image2_latentopt | IMAGE | Image 2 for reference latent (RGB only, expects correct size) | |
| image3_latentopt | IMAGE | Image 3 for reference latent (RGB only, expects correct size) |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| conditioning | CONDITIONING | Text+vision embeddings with reference latents metadata attached |
| latent | LATENT | Image1 latent in standard format (for VAEDecode or other latent nodes) |
| formatted_prompt | STRING | Final ChatML-formatted prompt with vision tokens (for debugging) |
| recommended_cfg | FLOAT | ⭐ NEW: Recommended CFG scale based on preset (2.5-5.5). Connect to KSampler's cfg parameter for optimal image/text balance! |