Noct Q Studio Demo

✨ Author Presets • click to load prompt & settings

ZeroGPU Optimized: INT8 convrot (7.3 GB)

Describe your image (Full sentence prose) 0 / 4000

Append Scene Details

Canvas Aspect Ratio

💡 1024×1536 is the model author's recommended native canvas for realistic full-body generations.

Inference Steps 25

Recommended: 25 steps with Euler Simple

Guidance (CFG) 3.0

3.0 follows author workflow; 1.0 ignores negative prompt

Seed Number

Randomize seed on each generation

Negative Prompt

Generation Output Ready

Running Noct Q V4 INT8...

Sampling with Euler 25 steps

⬇️ Download

Model Noct-Q-Uncensored-Qwen-Image-2.1

Checkpoint NoctQ_V4_int8_convrot.safetensors

Resolution 1024 × 1536

Execution Details 25 steps · CFG 3.0 · Euler

Official Model Showcase

High-fidelity sample generations from the author demonstrating lighting, realism, and prompt adherence.

1024 × 1536

Sunrise Lake Jetty

A photograph of a woman in her twenties with long wavy chestnut hair, wearing an oversized white linen shirt, sitting on a wooden jetty over a calm lake at sunrise with her bare feet in the water, holding a mug in both hands.

25 steps · CFG 3.0

1024 × 1536

Blue Hour Portrait

A cinematic atmospheric shot at dusk with soft cool ambient fill light and warm rim illumination. Detailed skin texture, crisp eyes, natural bokeh background.

25 steps · CFG 3.0

1024 × 1536

Direct Flash Editorial

High-contrast direct flash photography with natural skin pores, realistic shadow falloff, subtle film grain, editorial fashion realism.

25 steps · CFG 3.0

ComfyUI Execution Pipeline

Run the exact Noct Q V4 INT8 Convrot checkpoint locally or on cloud GPUs.

Node 01

Diffusion Model

Load NoctQ_V4_int8_convrot.safetensors (7.3 GB) via Checkpoint Loader.

Node 02

Qwen 3 VL CLIP

Load qwen3vl_8b_int8_convrot.safetensors for text encoding.

Node 03

KSampler (Euler)

25 steps, Euler sampler, simple scheduler, CFG 3.0, 1024x1536 empty latent.

Node 04

BF16 VAE Decode

Decode latent samples using qwen_image_2.1_vae_bf16.safetensors into PNG.

How to run in ComfyUI:

Ensure you are running ComfyUI 0.37 or newer (uses core nodes only, no external custom node requirements).
Download NoctQ_V4_int8_convrot.safetensors into models/diffusion_models/.
Download qwen3vl_8b_int8_convrot.safetensors into models/text_encoders/.
Download qwen_image_2.1_vae_bf16.safetensors into models/vae/.
Drag & drop NoctQ_V4_workflow.json directly onto the ComfyUI browser canvas.

Prompting & Generation Guide

How to get the cleanest realism and best adherence with Noct Q.

📝 Write Prose, Not Tags

Qwen Image 2.1 was trained on full-sentence natural captions, not comma-separated tag lists. Structure your prompt as a complete visual description:

[Shot type] + [Subject with explicit adult age] + [Action / Outfit / Pose] + [Environment / Setting] + [Lighting & Camera Details]

Example: "A photograph of a woman in her twenties with long wavy chestnut hair, sitting on a wooden jetty... soft golden light from the left."

⚡ Hardware & Efficiency

The INT8 convrot quantization allows this model to comfortably execute on consumer GPUs with 8 GB to 12 GB VRAM.

CFG 3.0: Follows the author's exact negative prompt guidance.
CFG 1.0: Runs nearly twice as fast and completely ignores the negative prompt.
25 Steps: Provides peak convergence with the Euler sampler.

License Notice:

Noct Q is built with Qwen. Qwen is licensed under the Qwen Research License Agreement. Free for non-commercial research use.