SenseNova U1.5 Lite vs Qwen-Image 3.0 vs Z-Image-Turbo tested in ComfyUI
Klutzy_Cap8492 · reddit · 2026-11-05
A Redditor tested three Chinese open-source image models — SenseNova U1.5 Lite, Qwen-Image 3.0, and Z-Image-Turbo — in ComfyUI across six tasks: annotated edits, four-reference composition, poster style adaptation, and text-heavy layouts.
Key findings:
- Garden edit: SenseNova removed markings as instructed; Qwen kept blue outlines.
- Four-reference composition: both SenseNova and Qwen preserved photo/anime style distinctions but missed details (open eyes vs. changed pose).
- Poster adaptation: SenseNova better preserved the reference layout; Qwen better for flexible redesigns.
- Infographic: Qwen cleanest; SenseNova duplicated copy, Z-Image-Turbo omitted a section.
- Scientific diagram: small labels and arrow relationships need separate verification.
Author's selection rule: test the least forgiving part of your actual task first. Full prompts and ComfyUI setup docs are included.
More from Multimodal
- Full Seedance 2.5 prompt breakdown: OUTPUT, STYLE and active references — socialwithaayan · 2026-11-05
- Roadrunner 3-panel character sheet prompt for cross-shot consistency — socialwithaayan · 2026-11-05
- One Seedance 2.5 prompt generates a 30-second 11-shot animated chase — socialwithaayan · 2026-11-05
- Ninada open-sources 48 in-browser 24kHz TTS models for Sanskrit and 10 Indic languages, 538ms on phone WASM — init0 · 2026-11-05
- NodeUI ships a native ComfyUI workflow editor for iPhone, iPad and Android at $4.99 — aguhl0614 · 2026-11-05
- Training a Z-Image Turbo LoRA from 24 photos: why 2,750 steps beat 3,000 — meshive · 2026-11-05