Isometric Redmond Z-Image LoRA

TL;DR

• LoRA adapter for Tongyi-MAI/Z-Image-Turbo that generates consistent isometric illustrations
• Trigger words: ISOMETRIC and ISOMETRICREDM
• Apache 2.0 license — commercial use permitted
• Best results with ComfyUI workflows

What This Model Does

Isometric Redmond Z-Image is a Low-Rank Adaptation (LoRA) trained on top of Tongyi-MAI's Z-Image-Turbo, a fast text-to-image diffusion model. The LoRA specializes in isometric illustration style — clean, geometric 3D-like projections with a fixed 30-degree viewing angle. Think architectural diagrams, game world maps, technical documentation art, and environment concept art.

The training data consisted of curated isometric illustration images. The result is a lightweight adapter (Safetensors format) that steers Z-Image-Turbo toward this specific aesthetic without retraining the base model.

Trigger Words & Usage

Include these tokens in your prompt to activate the style:

ISOMETRIC, ISOMETRICREDM

Example prompt:

ISOMETRIC, ISOMETRICREDM, isometric view of a modern co-working space, clean lines, minimal color palette, soft lighting

The model card recommends ComfyUI for the best results. Load the LoRA into your ComfyUI workflow (typical strength 0.7–1.0) and generate.

Recommended Settings

Parameter | Recommendation

|---|---

Base Model | Tongyi-MAI/Z-Image-Turbo LoRA Strength | 0.7 – 1.0 Sampler | Euler a / DPM++ 2M Karras Steps | 20 – 30 CFG Scale | 5 – 7 Resolution | 1024×1024 (native) Z-Image-Turbo is a distilled/fast model, so fewer steps work well. The LoRA was trained at 1024×1024; other resolutions may work but consistency is best at native size.

License

Apache License 2.0 — permissive, allows commercial use, modification, and distribution with attribution. See the HF model card for the full license text.

Examples

The three sample images above (from the model card) demonstrate the range: urban/architectural exteriors, interior spaces, and outdoor environments — all sharing the clean isometric aesthetic.

Common Errors and Fixes

Error | Cause | Fix | Source

|---|---|---|---

Style not activating | Missing trigger words | Add ISOMETRIC, ISOMETRICREDM to prompt | Model card Inconsistent perspective | LoRA strength too low | Increase LoRA weight to 0.8–1.0 | Model card / Blurry outputs | Too few steps on base model | Use 20+ steps, Euler a sampler | Z-Image-Turbo docs Wrong aspect ratio | Non-square generation | Generate at 1024×1024, crop post-hoc | Training resolution

FAQ

Q: Can I use this with Automatic1111 / WebUI Forge? A: Yes — any UI that supports loading Safetensors LoRAs for SDXL/flux-style models will work. ComfyUI is recommended by the author for best results.

Q: What base model version exactly? A: Tongyi-MAI/Z-Image-Turbo on Hugging Face. Ensure you load this exact base model; other "Turbo" variants are not compatible.

Q: Is this a full model or just a LoRA? A: LoRA only (~few hundred MB). You must have Z-Image-Turbo downloaded separately.

Q: Commercial use allowed? A: Yes, Apache 2.0 permits commercial use with license attribution.

Sources

• Hugging Face Model Card — model card, trigger words, license, sample images
• Z-Image-Turbo Base Model — base model reference
• Sample images inspected via vision: consistent isometric projection, clean geometric style, architectural/interior/environment subjects

Enriched for ArtificialGuyBR portfolio. Model trained with GPU support from Redmond.AI.