ThatPainter is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Learn More
Yes—FLUX.2 [klein] 9B can edit an existing photograph into a cinematic, illustrated, editorial, period, anime-inspired, or branded visual style. You do not need a LoRA for ordinary style changes. Use the fast distilled black-forest-labs/FLUX.2-klein-9B checkpoint for direct image-to-image editing; use black-forest-labs/FLUX.2-klein-base-9B when you want to train or run a custom LoRA for repeatable style, character, product, or domain consistency.
The important limitation is licensing: the 9B models use Black Forest Labs’ FLUX Non-Commercial License, so downloadable weights are not automatically cleared for commercial client work.
What FLUX.2 [klein] 9B actually does
Released on January 15, 2026, the FLUX.2 [klein] family is a 9-billion-parameter rectified-flow transformer designed for both text-to-image generation and image editing. Its editing workflow accepts a source image and a written instruction, and the model supports single-reference and multi-reference editing.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minute#1 Best Overall
- System Compatibility Note: This 2-slot card measures 271 x 112 x 39 mm and requires a single 12V-2x6-pin power connector. Please verify chassis and PSU compatibility before purchase.
- Dedicated Support: Please contact us directly through Amazon for any product questions or assistance you may require.
- Professional Intel Arc Pro B70 GPU: Built on the Intel Xe2-HPG architecture, it features 32 Xe cores and 256 XMX engines, designed to accelerate AI, rendering, and complex visualization workloads.
- Massive 32GB GDDR6 VRAM: Equipped with 32GB of high-speed GDDR6 memory on a 256-bit bus, running at 19 Gbps, which allows for handling large AI models and complex datasets locally.
- High-Performance Engine Clock: Delivers an engine clock of 2540 MHz, providing the compute power needed for demanding professional applications and AI inference.
That makes it a generative image editor rather than a pixel-perfect retouching tool. It can preserve the broad subject and composition while changing medium, palette, lighting, atmosphere, or visual language—but hands, small objects, facial details, product geometry, logos, and background elements may shift.
Black Forest Labs describes the distilled 9B checkpoint as a fast, production-oriented model designed for very few inference steps. The separate Base checkpoint retains the undistilled training signal and is the recommended starting point for LoRA training and customization. See the official FLUX.2 repository and the 9B model card.
Which model should you download?
| Goal | Checkpoint | Why |
|---|---|---|
| Fast everyday photo editing | black-forest-labs/FLUX.2-klein-9B |
Distilled for fast inference and direct editing. |
| Training a custom LoRA | black-forest-labs/FLUX.2-klein-base-9B |
Officially recommended for LoRA training, fine-tuning, research, and custom pipelines. |
| Lower memory use or a more permissive model license | FLUX.2 [klein] 4B | Black Forest Labs lists the 4B family under Apache 2.0 and with substantially lower VRAM requirements. |
Do not assume that an adapter made for 9B Base works with distilled 9B, 9B KV, 4B, or another FLUX checkpoint. Before loading a community LoRA, check its target architecture, trigger word, recommended strength, inference steps, and software requirements.
Hardware and software requirements
Hardware needs are substantial. Black Forest Labs’ model page lists these vendor estimates:
| Variant | Listed VRAM | Listed RTX 5090 inference time |
|---|---|---|
| 9B distilled | Approximately 19.6 GB | Approximately 2 seconds |
| 9B Base | Approximately 21.7 GB | Approximately 35 seconds |
| 4B distilled | Approximately 8.4 GB | Approximately 1.2 seconds |
| 4B Base | Approximately 9.2 GB | Approximately 17 seconds |
Separately, the Base model card says the model fits in approximately 29 GB of VRAM. These are not necessarily contradictory: memory depends on precision, resolution, quantization, text-encoder placement, and CPU offloading. Treat them as published estimates, not guaranteed thresholds. A 24 GB GPU may work with some configurations, but should not be promised as universally sufficient. Training a LoRA can require more memory than inference.
For Python and Diffusers, the distilled model card gives this general installation command:
Rank #2
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
pip install -U diffusers transformers accelerateFor Base, the current model card recommends the latest Diffusers source when necessary:
pip install git+https://github.com/huggingface/diffusers.gitLibrary APIs change. Check the current model card before installing, particularly if a pipeline class or argument differs from the example below. Hugging Face also requires accepting the applicable license and access conditions before downloading the 9B files.
Edit a photo without a LoRA
Start without an adapter. A LoRA is optional when you simply want to restyle a photograph. Install the dependencies, load a source image, describe what should change, and explicitly identify what must remain stable.
import torch
from diffusers import Flux2KleinPipeline
from diffusers.utils import load_image
device = "cuda"
dtype = torch.bfloat16
pipe = Flux2KleinPipeline.from_pretrained(
"black-forest-labs/FLUX.2-klein-base-9B",
torch_dtype=dtype,
)
pipe.enable_model_cpu_offload()
input_image = load_image("input.jpg")
image = pipe(
image=input_image,
prompt="Transform this photo into a cinematic oil painting",
).images[0]
image.save("edited.png")The example uses Base because it follows the current Base model card. For fast ordinary editing, substitute black-forest-labs/FLUX.2-klein-9B and follow that checkpoint’s current official example. Do not assume every Diffusers release accepts precisely the same pipeline class or parameters.
A prompt structure that works better than “make it artistic”
[action] + [subject preservation] + [target medium/style] +
[color and lighting] + [composition constraints] + [quality constraints]For example:
Transform the uploaded portrait into a hand-painted editorial gouache illustration.
Preserve the person's facial identity, pose, camera angle, hairstyle, clothing silhouette,
and background layout. Use muted teal, ochre, and warm cream colors, visible brush texture,
soft directional window light, and a refined magazine-illustration finish. Do not add text,
logos, extra people, or new accessories.Describe the style through medium, surface, palette, lighting, and composition instead of relying only on an artist’s name. Make one major change at a time. “Change the medium, pose, clothing, location, lighting, and identity” is much harder to control than “change this portrait into a gouache illustration while preserving the pose and face.”
Three useful starting prompts
- Cinematic portrait: “Transform this portrait into a cinematic still with low-key directional lighting, subtle film grain, restrained shadows, a warm skin-tone palette, and a shallow depth-of-field look. Preserve the face, pose, clothing, and camera angle. Add no text or logos.”
- Watercolor illustration: “Convert this photograph into a delicate watercolor painting with translucent washes, soft paper texture, simplified edges, and a cool spring palette. Keep the subject, silhouette, perspective, and main background shapes recognizable.”
- Retro editorial: “Restyle this photograph as a 1970s editorial print with muted ochre, olive, cream, and rust colors, analog grain, flat graphic shadows, and slightly faded ink. Preserve the subject’s pose and composition. Do not add typography or period accessories.”
Prompt following can fail, and the official model card warns that results depend on prompting style. Run several seeds and compare outputs rather than treating one generation as definitive.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Rank #3
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
What a LoRA adds
A LoRA is a small adapter that teaches a compatible base model a reusable visual concept. It is useful when prompting alone cannot produce the same result repeatedly:
- A studio’s recognizable illustration or painting language.
- A character or fictional mascot that must remain consistent.
- A product, garment, object, or visual catalog.
- A specialized domain such as a particular type of architecture or fashion.
- A repeatable transformation applied across many source photographs.
For a single portrait or a one-off watercolor conversion, try the base model first. Training a LoRA adds dataset preparation, experimentation, memory requirements, compatibility checks, and licensing responsibilities.
Train a FLUX.2 Klein LoRA
Use FLUX.2 [klein] 9B Base, not the fast distilled checkpoint, as your training starting point. Black Forest Labs’ training guide identifies style transfer, character consistency, domain specialization, and concept learning as suitable LoRA use cases.
- Choose one objective. Decide whether the adapter teaches a style, person, character, product, object, or domain. Avoid mixing unrelated objectives at the beginning.
- Collect legally usable images. Use varied, clean examples with the permissions needed for your intended use. For identity training, obtain appropriate consent.
- Caption consistently. Describe the subject, environment, pose, lighting, and relevant visual attributes. Use a unique trigger token for a subject or style.
- Keep the trigger distinct. Do not make the trigger synonymous with a generic attribute such as “portrait” or “blue.” It should identify the trained concept.
- Start with a controlled run. Exact dataset size, rank, learning rate, and step count are not universal FLUX.2 rules; follow the selected training toolkit’s guidance.
- Test held-out photographs. Use images that were not in training and compare multiple subjects, poses, lighting conditions, and backgrounds.
- Diagnose the result. Adjust image selection, captions, training duration, learning rate, rank, or adapter strength according to whether the adapter is underfitting or overfitting.
- Save and document the adapter. Record the base checkpoint, trigger word, training tool, software version, settings, dataset rights, and license.
Black Forest Labs says AI-Toolkit is optimized for consumer GPUs with 12 GB or more of VRAM, but that is a toolkit-level claim, not a guarantee that every 9B Base training configuration will fit. Gradient checkpointing, CPU offload, quantization, or a cloud GPU may be necessary.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteLoad and use an existing LoRA
The official training documentation gives this general loading pattern:
import torch
from diffusers import Flux2KleinPipeline
pipe = Flux2KleinPipeline.from_pretrained(
"black-forest-labs/FLUX.2-klein-base-9B",
torch_dtype=torch.bfloat16,
)
pipe.load_lora_weights("path/to/your_lora.safetensors")
pipe.to("cuda")
image = pipe(
"a photo of ohwx in a garden on a sunny day",
num_inference_steps=50,
).images[0]Here, ohwx is only an example. Replace it with the exact trigger supplied by the adapter creator. Use the creator’s recommended LoRA weight and the syntax supported by your installed Diffusers version; there is no universal best strength.
Rank #4
- System Compatibility Note: 2-slot card, 271x112x39mm, single 8-pin power, 200W TDP. Verify chassis clearance and PSU capacity before purchase.
- Dedicated Support: Please contact us directly through Amazon for any product questions or assistance you may require.
- 24GB GDDR6 on 192-Bit Bus: Massive 24GB memory with 456 GB/s bandwidth – ideal for LLMs, AI inference, 3D rendering, and generative design.
- Intel Xe2-HPG Architecture: Built on Intel's next-gen architecture with 20 Xe cores and 160 XMX engines for AI acceleration (197 INT8 TOPS).
- PCIe 5.0 Support: PCI Express 5.0 x16 interface for maximum bandwidth with the latest workstation platforms.
For image-to-image editing, combine the trigger with a preservation-focused instruction, for example: “Transform the uploaded portrait into the ohwx editorial gouache style. Preserve the person’s pose, clothing silhouette, camera angle, and background layout.” If the adapter is trained for a character rather than a style, expect it to affect identity more strongly.
Preserve the original photograph as much as possible
Tell the model separately what to preserve and what to replace. Useful preservation details include:
- Facial identity and approximate age.
- Pose, expression, framing, and camera angle.
- Hairstyle, clothing silhouette, and major accessories.
- Background layout and horizon line.
- Product dimensions, color blocks, and visible components.
- Lighting direction and overall composition.
Identity preservation is an objective, not a guarantee. A style adapter may alter facial structure, hair, skin detail, or age. A character adapter may preserve identity while reducing variation in pose and scene. Reference images can help where the workflow supports them, but compare several source photos and seeds.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Common failure modes
Wrong checkpoint
If an adapter targets 9B Base but you load distilled 9B, 4B, 9B KV, or another FLUX architecture, it may fail to load or produce poor results. Verify the target model before troubleshooting prompts.
Overfitting
Symptoms include every subject acquiring the same face, pose, background, or color treatment, or the model copying training images too literally. Use more varied examples, reduce training duration or LoRA strength, improve captions, and validate on held-out photographs.
Underfitting
If the trigger has little effect and the style remains weak, improve image quality and caption consistency, use a more distinctive trigger, or adjust training duration and learning rate according to the chosen toolkit’s instructions.
Recommended Free Tools
Best Value
- PLEASE NOTE: Exporting an NVIDIA RTX Pro 6000 GPU outside the US requires strict adherence to the U.S. Export Administration Regulations (EAR) and issuance of an export license from the Bureau of Industry and Security (BIS). Compliance and Know Your Customer (KYC) screening may be required as a condition of order acceptance. [NVIDIA Blackwell Streaming Multiprocessor] The new SM features increased processing throughput, and new neural shaders that integrate neural networks inside of programmable shaders | DLSS 4: Multi Frame Generation ensures ultra-smooth frame pacing for lifelike simulations.
- [Double-Flow-Through Design] The RTX PRO 6000 Blackwell features a double-flow-through cooling design, optimizing efficiency and airflow to sustain peak performance under 600W power loads. | [5th Gen Tensor Cores] Deliver up to 3X the performance of the previous generation and support for FP4 precision for faster AI model processing times with reduced memory usage, enabling local fine-tuning of LLMs and generative AI | [4th Gen Ray Tracing Cores] Double the ray-triangle intersection rate of the previous generation to create photoreal, physically accurate scenes and immersive 3D designs with RTX Mega Geometry, which enables up to 100X more ray-traced triangles.
- [PCIe Gen 5] Support for PCIe Gen 5 provides double the bandwidth of PCIe Gen 4, improving data-transfer speeds from CPU memory and unlocking faster performance for data-intensive tasks like AI, data science, and 3D modeling. | [GDDR7 Memory] With 96 GB of GPU memory and 1.8 TB ps bandwidth, it can tackle massive 3D and AI projects, fine-tune AI models locally, explore large-scale VR environments, and drive larger multi-app workflows.
- [DisplayPort 2.1] Achieve unparalleled visual clarity and performance, driving high resolution displays at up to 8K at 240 Hz and 16K at 60 Hz. Increased bandwidth enables seamless multi-monitor setups while HDR and higher color depth support ensures superior color accuracy for precision work, such as video editing, 3D design, and live broadcasting.
- [Universal MIG] Divide a single RTX PRO 6000 Blackwell into multiple isolated instances, each with dedicated resources, allowing for concurrent execution of multiple workloads, optimized GPU utilization, and secure isolation of different applications or users. [WARRANTY] 3 YR Manufacturer's Warranty. Bulk OEM Packaging. Retail Packaging is NOT included.
Text, labels, and logos
The model card warns that rendered text can be inaccurate or distorted. Do not rely on this workflow for final logos, packaging copy, legal notices, labels, or signage without manual correction.
Composition drift
Even when the overall scene survives, hands, jewelry, fine patterns, small objects, product geometry, and background details can change. Use a conventional retouching or compositing workflow when exact pixels matter.
Local, hosted, or 4B?
| Choose | Best for | Main trade-off |
|---|---|---|
| Local 9B | Privacy, automation, custom graphs, and full workflow control. | High VRAM, storage, setup complexity, and restrictive 9B licensing. |
| Black Forest Labs Playground | Trying the model without installing a GPU workflow. | Check current access, billing, privacy, and retention terms. |
| Black Forest Labs API | Applications, batch jobs, and hosted production pipelines. | Images go to a third party and costs scale with usage and megapixels. |
| Local 4B | Smaller GPUs, lower latency, and investigation of the Apache 2.0 option. | Lower capacity than 9B and still requires license review for dependencies and adapters. |
Black Forest Labs’ current pricing documentation says FLUX.2 Klein 9B editing starts at approximately $0.015 and scales by megapixel, with one credit equal to $0.01. Pricing and Playground access can change, so check the official pricing documentation before committing to a hosted workflow.
For node-based local work, the official repository says Klein models are available in ComfyUI. ComfyUI is a good choice for reusable graphs, references, masks, and adapters, but third-party nodes and workflows are not automatically validated by Black Forest Labs. Python developers may prefer Diffusers for scripts, notebooks, and custom applications.
Free tools Windows power users keep installed
One-click scans. No signup required.
License, privacy, and responsible use
Before commercial use, review all of these separately:
- The current FLUX.2 9B model license and Acceptable Use Policy.
- The LoRA’s own license and any restrictions from its creator.
- Rights to the training photographs and output source images.
- Consent for identifiable people and client-owned material.
- Terms, retention, privacy, and commercial rights for any hosted API or Playground.
- Restrictions on harmful, deceptive, privacy-invasive, or non-consensual uses.
Local execution does not make commercial use automatically legal. Likewise, a generated output is not automatically cleared simply because the model files were downloadable. The official model card documents both access requirements and important limitations, including prompt-following failures and biases inherited from training data.
Final recommendation
Start with distilled FLUX.2 [klein] 9B if your goal is to restyle existing photographs quickly. Use a structured prompt that names the transformation and protects the subject, pose, composition, and important details. Add a LoRA only when you need a look, identity, product, or domain concept to recur consistently. If you train one, use 9B Base, document its compatibility, and test it on photographs outside the training set.
Choose 4B when hardware, speed, or the Apache 2.0 model license is more important than the additional capacity of 9B. In every case, treat “any photo” and “any look” as creative shorthand—not a promise of arbitrary style control, perfect identity preservation, accurate typography, or unrestricted commercial rights.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
