🎨 Hrio Design｜三方案并发

Hrio_Design_Three_View_Node (🎨 Hrio Design｜三方案并发) is the manual version of the pack's template panel. The template node hands you a dropdown and generates three proposals from a preset; this node hands you three prompt boxes - front, side, back - and does the same concurrent generation so you can compare three takes on the same brief side by side. If you already have your own three angles in mind, this is the node you reach for, because it doesn't make you fight the template system.

It's still a cloud node, full stop. The three prompts get packed off to the author's private Gemini-format backend (banano2, banano-pro, veo3.1, or gemini3.1-pro as the model), reference images get uploaded once and reused across all three requests, and three IMAGE tensors come back. No GPU, no local weights, API key required, per-call metered. And because it runs three concurrent paid jobs per execution, the cost math matters more here than on the single-image node.

How it works

The three concurrent requests share the same uploaded reference images (the image_1–image_10 tooltips say the same batch is reused by all three concurrent requests). Each view is generated, cached, and - like the template node - can be retried or regenerated on its own. The pack also hands you an automation_payload JSON field and a web automation subsystem for batch runs, but the ordinary use is dead simple: write three prompts, hit run, compare.

The inputs that matter

• front_prompt, side_prompt, back_prompt - the three proposal briefs. The defaults are nicely differentiated: 方案 A is the formal main design, 方案 B pushes atmosphere and material detail, 方案 C goes more experimental while staying "high-end and grounded." Keep that A/B/C structure and you'll get genuinely different takes instead of three reskins.
• model - banano2 (default) is the pack's own hosted image model; banano-pro, veo3.1, and gemini3.1-pro are alternates, with the last two being Google's closed models routed through the author's backend.
• api_key - blank falls back to config.ini's [banana] api_key.
• image_size - 1K / 2K / 4K / 8K, default 2K, shared by all three images. aspect_ratio - Auto plus presets, default Auto, shared across the three views.
• generate_scope - regenerate all three, or just front/side/back. Regenerating one view reuses the cached results of the other two, so you're not billed three times to fix one image. This is the node's best feature.

Outputs

• front_image, side_image, back_image - three IMAGE outputs, one per prompt. Stack them on a preview node and you get your side-by-side comparison.
• images - an IMAGE output combining the three, for single-stream downstream use.
• info - STRING run metadata. mp4url - STRING URL, for when the node's result set is turned into a video.

Installing it

Same pack as the rest of Hrio Design. ComfyUI Manager search Hrio-Design, or:

cd ComfyUI/custom_nodes
git clone https://github.com/moki-flank/Hrio-Design

Restart ComfyUI. Only requests, numpy, and pillow are needed; no model downloads, since inference never happens locally.

Where people get burned

The cost trap is the same one as the template node, just easier to miss because the inputs look like a normal prompt: every run is three billed generations, and generate_scope's one-view regeneration exists precisely so you don't re-pay for the two you're happy with. Also worth knowing: the pack calls these prompts "front/side/back" but they're really just variant A/B/C - don't expect an actual multi-angle 3D reconstruction, it's three independent proposals. And the usual pack caveats apply: verify_ssl = false in config.ini, a key that only the author issues, and no community troubleshooting to lean on if a run fails.