Skip to content

FLUX Klein 9B Inference

★★★★★ Intermediate

Practical reference for FLUX.2 Klein 9B image generation. Covers optimal sampler settings, multi-pass upscaling, LoRA stacking, anatomy fixes, and prompting patterns.

Distilled 9B (4-Step Model)

Parameter Value Notes
Steps 4 Designed for 4-step inference
CFG 1.0 Official recommendation. Never exceed 2.0
Sampler euler Most stable for edges and text sharpness
Scheduler simple Alternative: Flux2Scheduler (resolution-aware)
Denoise 1.0 For txt2img

CFG above 2.0 produces "deep-fried" artifacts - the distilled model has guidance baked in. Using 50+ steps wastes compute and actively degrades quality.

Anatomy Fix: Use 8 Steps

At 4 steps, complex poses (seated, two-person) often produce extra limbs. 8 steps shows dramatic improvement for anatomy without excessive slowdown. Beyond 8 steps on distilled: diminishing returns.

Base 9B (Non-Distilled)

Parameter Value Notes
Steps 20-24 Sweet spot. Below 15 loses detail.
CFG 3.5-5.0 5.0 gives best prompt adherence
Sampler euler Most stable overall
Scheduler simple sgm_uniform for upscale passes
Denoise 1.0 For txt2img

Why Base Over Distilled

  • Higher output diversity
  • Configurable quality/speed tradeoff
  • Better for LoRA training and custom pipelines
  • Better input for downstream upscalers

Sampler Deep-Dive

Sampler Scheduler Use Case
euler simple Default gold standard
euler Flux2Scheduler Resolution-aware, adapts to aspect ratio
res_2s simple Better anatomy (2x compute/step)
res_2m ddim_uniform High quality general purpose
dpmpp_2m_ancestral sgm_uniform Analog grain, cinematic/film texture
euler_ancestral_cfg++ sgm_uniform Detail enhancement in upscale passes

Key insight: res_2s at 4 steps equals euler at 8 steps in compute. Fixes anatomy without increasing step count explicitly.

Multi-Pass Upscaling

Base (Stage 1) + Distilled (Stage 2)

[Prompt] -> Klein Base 9B, 1024x1024, 20 steps, CFG 5.0, euler/simple
         -> Upscale Latent 2x -> 2048x2048
         -> Klein Distilled 9B, 4-8 steps, CFG 1.0, euler/simple, denoise 0.4-0.6
         -> VAE Decode -> 2048x2048 output

Fast Pipeline (Distilled Only)

[Prompt] -> Klein Distilled 9B, 1024x1024, 8 steps, CFG 1.0, euler/simple
         -> Image Resize -> 2048 longest side
         -> Klein Distilled upscale, 4 steps, euler_ancestral_cfg++,
            sgm_uniform, denoise 0.8, s_noise 1.2
         -> 2048x2048 output

Denoise Values for Second Pass

Goal Denoise Effect
Tight fidelity 0.3-0.4 Structure lock, minimal change
Balanced 0.5-0.6 Good detail, some creativity
Creative upscale 0.7-0.8 More prompt-driven
Detail enhancement 0.8 Best for realistic fine detail

4K+ Output

Add tiled upscale stage: 4x4 grid, 256px tiles with 128px overlap, seam_fix_denoise=1.0, seam_fix_width=32-128px. Florence2 auto-caption before upscale reduces hallucinations at tile borders.

Seam Artifacts

Causes: tile boundary context loss, latent space discontinuity, high denoise in second pass.

Fixes: 128px overlap padding, mask blur 12-16px, half-tile reprocessing at boundaries, band-pass filtering (retain high-freq detail, blend low-freq).

LoRA Stacking

Capacity and Strength

Up to 3 LoRAs simultaneously, each with individual weight (0-4).

Strength Effect
0.0-0.3 Subtle, barely visible
0.4-0.75 Sweet spot - balanced texture + coherence
0.73 Recommended default for single LoRA
0.8-1.0 Maximum texture, starts pulling apart
1.0+ Coherence loss, visible artifacts

Multi-LoRA Rules

  • Stack strongest-influence LoRA first (processed sequentially)
  • Total combined strength should stay under 1.5-2.0
  • If artifacts appear, reduce weakest LoRA first
  • Watch for style conflicts (two different color grading LoRAs)

Klein vs Dev LoRA Behavior

Same LoRA settings produce fundamentally different results: - Klein: heavier grain structure (16mm film look) - Dev: cleaner, more like 35mm film - FP8 Klein: maintains desirable grain; non-FP8 is cleaner

Anatomy Problem Hierarchy

From most to least effective:

  1. Increase steps (4 -> 8): strongest lever for complex poses
  2. Simplify pose: standing > seated, single > multi-person, front > twisted
  3. Adjust CFG carefully: 1.0 baseline, 1.2 can fix fused fingers, >1.5 risks new problems
  4. Use res_2s sampler: doubles compute per step, fixes anatomy implicitly
  5. Negative prompting: "distorted features, unnatural proportions, extra limbs"
  6. Use Base model: 20 steps rarely has anatomy issues (much slower)

Face Blur Fix

  • Increase steps (most effective)
  • Higher resolution (1536x1920 for portraits)
  • Flux2Klein-Enhancer node to boost text conditioning magnitude

Flux2Klein-Enhancer Node

Custom node for stronger prompt adherence:

Parameter Range Default Purpose
Magnitude 0.0-3.0 1.0 Text embedding scaling
Contrast -1.0-2.0 0.0 Token difference amplification
Normalize Strength 0.0-1.0 0.0 Token magnitude equalization
Edit Weight 0.0-3.0 1.0 Preservation vs prompt following
Ref Strength 0.0-5.0 1.0 Reference structure lock (0=txt2img)
Blend with Noise 0.0-1.0 0.0 Reference-noise interpolation

Dampening 1.20-1.30 recommended for precise preservation.

Prompting Guide

Structure

Subject -> Setting -> Details -> Lighting -> Atmosphere

Write as flowing prose, not keyword lists. Front-load important elements.

Lighting (Highest Impact)

Specify: source type, quality, direction, temperature, surface interaction. "Soft, diffused natural light filtering through sheer curtains" >> "good lighting"

Prompt Length

Length Words Use
Short 10-30 Concept exploration
Medium 30-80 Production work
Long 80-300+ Complex editorial/product shots

Style Annotations (End of Prompt)

  • "Style: Country chic meets luxury lifestyle editorial"
  • "Shot on 35mm film with shallow depth of field"
  • "Mood: Serene, romantic, grounded"

Native Resolutions

Both 4B and 9B support 11 aspect ratios up to 4MP (2048x2048 square). Range from 1:1 to 21:9.

Gotchas

  • 4B LoRA incompatible with 9B: different text encoder sizes (Qwen 3-4B vs Qwen 3-8B). LoRAs are not interchangeable between Klein 4B and Klein 9B.
  • FP8 specifically benefits from 8 steps: the FP8 quantized distilled model needs more steps than bf16 to match quality. Budget 8 steps minimum for FP8 inference.
  • Qwen-Image VAE artifacts: the VAE decoder can introduce washed-out details and checkerboard noise. A dedicated fix LoRA exists (strength 1.0, trigger: "Remove compression artifacts. Restore the fine details of the photo."). Only works on Qwen-VAE artifacts - will degrade images from other VAEs.

See Also

No cookies. No tracking scripts. Privacy-friendly analytics by Cloudflare.