What it does. This addon turns Blender into a full AI image generation studio. Pick a model from the built-in browser, stack LoRAs, type a prompt, hit generate. Your image appears right in Blender — UV-mapped onto a new plane, assigned to your active object, or saved to disk. All through a clean 7-step guided sidebar, no ComfyUI node editor required.
How it works. The addon connects to a free, locally running copy of ComfyUI over HTTP. ComfyUI runs the actual image generation on your GPU. Vision-Air handles the entire workflow automatically — model loading, LoRA chaining, sampling, and image delivery. Generation runs in a background thread so Blender stays fully interactive while your image is being made.
The only setup: Download ComfyUI (free, open source), start it, leave it running. The addon connects automatically. You never open the ComfyUI web interface, never build a node graph, never touch a workflow JSON file. Just open Blender, and generate.
The 7-step guided workflow
| Step | Panel | What you do |
|---|
| 1 | Select Model | Pick from the browser — settings and prompt tags auto-apply. Direct download link shown for each required file. |
| 2 | LoRAs | Add unlimited LoRAs from your ComfyUI library. Each slot has a strength slider (−2.0 to +2.0). Toggle on/off without removing. Global strength multiplier. Warning if you stack too many for the arch. |
| 3 | Prompt Enhancer | Type a rough idea. A local text encoder rewrites it into a detailed prompt, which lands in Step 4 automatically. (Optional) |
| 4 | Prompt | Positive + negative prompts. Model tag hints shown inline — one click inserts the recommended prefix for the chosen model. |
| 5 | Settings | 7 aspect ratios, 64–4096px resolution, 14 samplers, 7 schedulers, steps, CFG, seed (manual / random / locked). Live pixel preview. One-click resolution presets. |
| 6 | Generate | Batch 1–16 images per click. Live progress bar. Generates onto a new UV-mapped plane and/or applies to active object. LoRA summary visible before you generate. |
| 7 | Export | Auto-save with custom filename template ({model}, {seed}, {date}, {prompt}, etc). Copy image path to clipboard. Save individual results. |
362 Curated LoRAs — 6 architectures
Every LoRA in the recommended browser includes a direct download link, recommended strength, and trigger words. No CivitAI hunting. Filter by category, star your favourites, expand trigger-word hints inline.
🔥 Alright, we see you. For the artistically adventurous... 🔥
Look — life is short, your GPU is beefy, and who are we to judge? Tucked right inside the LoRA browser (behind a tasteful opt-in toggle, we're not animals) are 157 hand-curated NSFW LoRAs spanning every architecture. We're talking everything from tasteful boudoir all the way to "definitely not sharing your screen in that Zoom call." All clearly labeled, all with strength recommendations, all one download link away. You're welcome. We'll keep this between us. 😏
12 AI checkpoint models — with download links
// SDXL Photorealism
Juggernaut XL v97.1 GB · Photo
Juggernaut XL v107.1 GB · Photo
RealVisXL V5.06.9 GB · Photo
DreamShaper XL Turbo6.9 GB · 4–8 step
// Anime / Illustrious / Pony
Illustrious XL v0.16.9 GB · Anime
NoobAI XL Eps v1.17.1 GB · Anime
WAI-NSFW-Illustrious v146.9 GB · Anime
Pony Diffusion V6 XL6.5 GB · Pony
Animagine XL 3.16.9 GB · Anime
// Flux
FLUX.1 Dev fp8Recommended
FLUX.1 Dev fp1623.8 GB · Full
FLUX.1 Schnell fp84-step · Apache
Everything else in the box
- Model browser auto-applies settings — picks a model, instantly sets the right steps, CFG, sampler, scheduler, resolution, and prompt tags
- Unlimited LoRA stacking — add as many LoRAs as you want per generation, each with individual enable/disable toggle and per-slot strength slider
- AI Prompt Enhancer — rough idea in, detailed prompt out, using a local text encoder you point the addon at
- Batch generation (1–16) — queue multiple images in one click, each with a fresh seed
- Preset system — save and reload your entire setup (model + LoRAs + prompt + settings) with a name. Presets survive across .blend files
- History panel — up to 20 entries with thumbnail previews, timestamps, full metadata (model, seed, steps, CFG, resolution, prompt). Stored as base64 — survives Blender restarts. Recall prompt, re-export, or delete any entry
- Generate Plane — auto-creates a UV-mapped plane at the 3D cursor, correctly sized to your aspect ratio, with the image applied as a Principled BSDF material
- Apply to Active Object — assigns the generated image to your currently selected mesh's material
- 7 aspect ratios — 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9 with live output pixel dimensions
- 14 samplers — Euler, Euler A, ER SDE, Heun, Heun++ 2, DPM 2, DPM++ 2M SDE, DPM++ SDE, DPM Fast, and more
- 7 schedulers — Normal, Karras, Exponential, SGM Uniform, Simple, DDIM Uniform, Beta
- Seed control — manual, random, or lock. Reuse Last Seed button always visible after a generation
- 3 VAEs, 3 upscalers, 2 negative embeddings — all linked in the model browser with direct download links
- Custom filename templates — auto-named exports with {model}, {seed}, {steps}, {cfg}, {date}, {time}, {prompt} tokens
- Flux-specific workflow — automatically loads 4 separate files (UNet + CLIP-L + T5-XXL + VAE) and handles FluxGuidance node automatically
- 157 NSFW LoRAs — opt-in toggle in the LoRA panel. Clearly labeled, strength-rated, and ready to download. For the bold. 😏
The math doesn't lie. Midjourney charges $10/month — that's $120/year. Vision-Air is $5, once, forever. If you generate even a single batch of images a week, you'll have gotten more value out of this in the first month than a full year of Midjourney. And after that? Still $5.
