ArchAi3D_Qwen_Person_Perspective
Low angle for power, high angle for vulnerability — perspective is psychology
- perspective_prompt
- full_prompt
- preservation_hints
- system_prompt
Camera angle on a person is a statement. Low angle reads power, high angle reads vulnerability, and ArchAi3D_Qwen_Person_Perspective is the pack's node that turns that psychology into a Qwen-Edit prompt. It's the one the README singles out with its own guide, built specifically for re-shooting people from a new camera position while keeping the person the same - the thing generic camera nodes get wrong, because Qwen will happily redesign your subject along with the angle.
The core design decision is identity_preservation, which defaults to strict and expands into explicit instructions ("keep face, clothes, hairstyle, pose identical"). That's the honest recognition that Qwen-Edit's whole-frame regeneration will change your subject unless you fight it with language. There are also psychological_effect (vulnerability, power, intimacy…) and focal_point (which body part the shot emphasizes) - the difference between "a low angle" and "a low angle that makes this person look heroic."
How it works
Six perspective presets (eye_level_front by default, plus high/low/worm's-eye, side profile, three-quarter, dutch angle) each map to a researched angle phrase plus the right psychological framing. body_proportion guides the model on exaggeration (elongate for low angles, diminish for high), and three booleans - background_adaptation, lighting_reinforcement, composition_centering - each append a clause that keeps the scene, light and framing coherent with the new perspective. They all default to true and all read like they should.
Outputs: perspective_prompt and full_prompt (with scene context), a preservation_hints string carrying the identity instructions, and system_prompt for the encoder's system input.
The inputs that matter
- perspective_preset - pick the angle; it sets focal point and psychology defaults.
- identity_preservation -
strictis the recommendation for a reason. Loose/none exists but you'll regret it on anything client-facing. - scene_context - the scene the person is in ("in modern office", "outdoor garden"). Feeds both prompts.
- custom_additions - a free-text escape hatch for anything the presets miss.
Installing it
Ships in ComfyUI-ArchAi3d-Qwen by Amir Ferdos (ArchAi3d):
# ComfyUI Manager: search "ArchAi3d Qwen"
cd ComfyUI/custom_nodes
git clone https://github.com/amir84ferdos/ComfyUI-ArchAi3d-Qwen.git
cd ComfyUI-ArchAi3d-Qwen
pip install -r requirements.txt
Restart. Load a Qwen-Image-Edit model (2509/2511, GGUF-quantized in practice) and the pack's encoder, and wire the two prompt outputs in.
Common issues
- Strict isn't literal.
identity_preservation: strictis a strong instruction, not a face-lock. The community's consistent finding - documented in the identity-preservation KB - is that Qwen-Edit drifts most on faces it hasn't seen in training. For real faces, plan an inpaint pass afterward. - The model re-poses instead of re-shooting. A "camera perspective change" can come out as the person standing differently. Keep the subject centered and the identity clause connected.
- Chain drift. Every pass re-emits the frame, so a perspective change is also a subtle redesign of everything else. Background/lighting toggles help, but pixel-exactness isn't on the table with this model.
This is the node you reach for when you want a portrait re-shot at a storytelling angle - the psychological framing is a genuinely thoughtful touch that generic camera nodes don't give you.
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| perspective_preset | COMBO | eye_level_front | Camera perspective preset optimized for people (high angle = vulnerability, low angle = power) |
| identity_preservation | COMBO | strict | How strictly to preserve person's identity (strict = keep face/clothes/pose identical) |
| psychological_effect | COMBO | none | Psychological effect to convey (vulnerability, power, intimacy, etc.) |
| focal_point | COMBO | auto | Which body part to emphasize (auto = preset decides) |
| body_proportion | COMBO | natural | Body proportion guidance (exaggerate height for low angles, diminish for high angles) |
| scene_context | STRING | Optional: Scene description (e.g., 'in modern office', 'outdoor garden') | |
| background_adaptation | BOOLEAN | true | Ensure background elements also change to complement the new perspective |
| lighting_reinforcement | BOOLEAN | true | Ensure lighting reinforces the perspective effect |
| composition_centering | BOOLEAN | true | Keep subject centered in composition (recommended for best results) |
| detail_showcase | COMBO | high | Level of detail to showcase in focal areas |
| prompt_style | COMBO | balanced | Prompt verbosity (detailed = most control, concise = faster) |
| custom_additions | STRING | Optional: Additional custom instructions to append | |
| debug_mode | BOOLEAN | false | Print generated prompt to console |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| perspective_prompt | STRING | Generated perspective control prompt |
| full_prompt | STRING | Prompt with scene context |
| preservation_hints | STRING | Identity preservation instructions |
| system_prompt | STRING | ⭐ NEW v4.0: Perfect system prompt for this perspective! Connect to encoder's system_prompt input for optimal results. |