Many powerful image models can generate stunning visuals, but they are often difficult to customize for specific creative needs. Released by KreaAI, Krea 2 is a 12-billion-parameter open source text-to-image diffusion model designed to generate images directly from natural-language descriptions.
The model is built for a wide range of applications, including-Creative exploration,Commercial design workflows , Visual production, Developer integrations ,Research experiments. Instead of being only a generation model, Krea 2 is designed as a flexible foundation that can fit into different creative ecosystems. The detailed insights can be found be accessing their research paper.
Basically Krea team released two variants-
(a) Krea2 Raw (Base)-The Raw checkpoint is designed primarily as a foundation model rather than a final image-generation model. It is not recommended for direct inference/image gen use because its main purpose is customization and further training. Train your loras on the top of it and use with Turbo variant. This makes it useful for- Fine-tuning, Domain-specific training, Creating custom LoRAs, Research experiments.
 |
| Krea2 Raw/Base Showcase |
(b) Krea 2 Turbo- It is an 8-step distilled checkpoint built for fast, high-quality text-to image generation. The model is distilled from a fully post trained Reinforcement Learning (RL) checkpoint, allowing it to produce better-quality images compared to Raw while maintaining much faster generation. The trade-off is that Turbo produces less variation between outputs, making it more consistent but slightly less flexible.
 |
| Krea2 Turbo Showcase |
Installation
1. First, install ComfyUI if you are a new user. Older user required to do the ComfyUI update from the Manager by selecting Update ComfyUI option.
2. All Krea2 base and Turbo (Bf16/fp8/Nvfp4/Int8) model listed below from community. Choose any of the model for your use case and system resources. Save any of them inside ComfyUI/models/diffusion_models folder.
These are Krea2 Model variants repacked officially by ComfyUI Team:
| Sl No. | Model | Format | Description (VRAM Support) |
|---|---|---|---|
| 1 | krea2_raw_bf16.safetensors | BF16 | Full precision Krea 2 Raw model. Provides maximum image quality and accuracy. Requires higher VRAM, suitable for GPUs with 16GB+ VRAM. |
| 2 | krea2_raw_fp8_scaled.safetensors | FP8 Scaled | Quantized Krea 2 Raw version with reduced memory usage while maintaining good quality. Suitable for GPUs with around 8GB–12GB VRAM. |
| 3 | krea2_turbo_bf16.safetensors | BF16 | Turbo variant optimized for faster generation with BF16 precision. Requires higher VRAM, recommended for GPUs with 16GB+ VRAM. |
| 4 | krea2_turbo_fp8_scaled.safetensors | FP8 Scaled | Memory-efficient Turbo model using FP8 scaling. Designed for faster inference on mid-range GPUs with approximately 8GB–12GB VRAM. |
| 5 | krea2_turbo_mxfp8.safetensors | MXFP8 | Mixed FP8 quantized Turbo model optimized for efficiency. Suitable for GPUs with lower VRAM requirements, generally around 8GB+ VRAM. |
| 6 | krea2_turbo_nvfp4.safetensors | NVFP4 | Highly compressed Turbo model using NVIDIA FP4 format. Designed for low VRAM usage, suitable for GPUs with approximately 6GB–8GB VRAM. |
These are Krea2 variants quantized by developer Winnougan:
| Sl No. | Model | Format | Description (VRAM Support) |
|---|---|---|---|
| 1 | krea2_base_fp8.safetensors | FP8 Tensorwise | Recommended for NVIDIA RTX 40xx series GPUs. |
| 2 | krea2_base_mxfp8.safetensors | MXFP8 | Recommended for NVIDIA RTX 50xx Blackwell GPUs. |
| 3 | krea2_base_nvfp4.safetensors | NVFP4 | Optimized for RTX 50xx Blackwell GPUs only. |
| 4 | krea2_base_int8.safetensors | INT8 Row-wise | Recommended for NVIDIA RTX 30xx series GPUs. |
| 5 | krea2_base_convrot_int8.safetensors | INT8 + ConvRot | Best quality INT8 version, recommended for RTX 30xx GPUs. |
| 6 | krea2_turbo_fp8.safetensors | FP8 Tensorwise | Recommended for NVIDIA RTX 40xx series GPUs. |
| 7 | krea2_turbo_mxfp8.safetensors | MXFP8 | Recommended for NVIDIA RTX 50xx Blackwell GPUs. |
| 8 | krea2_turbo_nvfp4.safetensors | NVFP4 | Optimized for RTX 50xx Blackwell GPUs only. |
| 9 | krea2_turbo_int8.safetensors | INT8 Row-wise | Recommended for NVIDIA RTX 30xx series GPUs. |
| 10 | krea2_turbo_convrot_int8.safetensors | INT8 + ConvRot | Best quality INT8 version, recommended for RTX 30xx GPUs. |
3. Download text encoder (qwen3vl_4b_bf16.safetensors / qwen3vl_4b_fp8_scaled.safetensors). Choose any of them. Save this into ComfyUI/models/text_encoders folder.
4. Download Vae (qwen_image_vae.safetensors) and save this into ComfyUI/models/vae folder.
5. This is optional, if want to generate stylized image. Download Krea 2 Lora models released by the community, tested and listed by us. Save this into ComfyUI/models/loras folder.
6. Refresh and restart comfyui to take effect.
Workflow
1. Download Krea 2 raw/Turbo workflow (Krea2-basic.json) from our hugging face repository. Use same workflow for Krea2 Raw/Turbo model.
2. Drag and drop the workflow into ComfyUI.
3. Load the models(Krea2 Raw/Turbo, text encoders, vae) into its relevant nodes. Additional, if you want to use any of the Krea 2 lora models into the workflow, just search for Load Lora node and connect it between Load diffusion models node and KSampler node.
4. Set KSampler configuration -
Krea2 Raw/Base:
Resolution- up to 1024
Steps-52
CFG- 3.5-4.5
Sampler: er_sde
Scheduler- simple
Krea2 Turbo:
Resolution-1024 to 2048
Steps-8
CFG- 0 (disabled)
Sampler: er_sde
Scheduler- simple
5. Put positive prompts into prompt box. use natural prompting with detailed style to get the best out of it.
6. Hit run to start image generation.
Some of the generations with Krea 2 Raw and Turbo. The results are not cherry picked.
Prompt:
90s grunge-style magazine poster with a muted yellow and teal distressed paper background. Central subject: An American teenage girl casually leaning against a chain-link fence holding a worn skateboard, with braided over-the-shoulder hair under a backward navy baseball cap, wearing an oversized red tshirt reading 'SKATE OR DIE', a denim hot pants. Raw polaroid flash lighting casting harsh shadows. Layout details: Top-left has thick teal 'RIDE' text over jagged black 'FREE' text. Top-right features a sticker showing a retro black skateboard silhouette, text 'VOL 04', a red circular postmark stamp reading '1994', and a jagged box reading 'NO FEAR... 90S RULES' next to a small white urethane wheel icon. Middle-left has a masking tape-label with handwritten 'RADICAL', bold text '[Extreme] pure adrenaline', a neon pink 'STREET LEGEND' tag, a rough 'CONCRETE JUNGLE' box, a bright yellow star sticker, and a faded grey boombox graphic with dual cassette decks and an extended antenna in the background. Middle-right has 'OWN THE STREET / RAMP MODE' overlaid on black spray-painted graffiti shapes like dripping arrows and a small crown. Bottom-left features vertical English text 'KEEP PUSHING / NEVER STOP' in stencil font. Bottom-right has 'RECORDING / TAPE 1' and bold 'STREET KINGS' with a cracked white skull icon next to it. Retro grunge streetwear aesthetic.
Prompt:
A handmade scrapbooking-style corgi cafe staff introduction poster, photographed behind glass, portrait orientation. The background is plain kraft paper. At the top center, a white cutout banner has hand-drawn black English calligraphy 'Honorary Tellers', bracketed by cursive English 'Our Stars~' above and 'Staff' below, accompanied by bone and dog head doodles. Below the banner is a horizontal row of four cutout photographic corgi heads. First, a classic red-and-white corgi with a white hand-drawn speech bubble cutout containing handwritten English text reading 'Duo Duo: Big foodie, will let you pet for treats!'. Second, a tricolor corgi with a speech bubble reading 'Qiu Qiu: Energetic little dynamo, loves playing fetch!'. Third, a fawn corgi with a speech bubble reading 'Dou Dou: Patrol guard, always monitoring his territory!'. Fourth, a blue merle corgi with a speech bubble reading 'Quan Quan: Shy baby, please treat gently!'. The central section has a prominent warning reading 'They are gaining weight, please do not feed!' with green underlines and more dog doodles. To the left is a larger dog-head sketch and 'Dear Guests: Welcome!'. In the lower right of the middle section is a fifth dog cutout (a fluffy corgi) next to a large speech bubble reading 'Hi~ New Intern~ My name is Mao Mao, I am still training, please give me a hug!'. The bottom third of the poster is filled with a paragraph of handwritten black English text with green underlines, reading 'Thank you for visiting the Dog House! Our pups here are very enthusiastic and tend to jump on people, so please protect your knees! For the health of our fur babies, outside treats are strictly prohibited. Have a wonderful time!'. Small hand-drawn stars, bones, and paw prints dot the bottom. The overall style is charming, DIY cafe signage with slightly glossy paper cutouts pasted onto coarse brown paper.
Prompt:
A futuristic cyberpunk fashion editorial photograph of a female model standing confidently in a neon-lit studio, wearing a glossy black patent leather cropped jacket, metallic vinyl mini skirt, black fishnet stockings, and extreme platform thigh-high boots. The outfit has a rebellious futuristic nightclub aesthetic with reflective synthetic materials, sharp silhouettes, and high-fashion cyber styling. The model has sleek dark straight hair, dramatic futuristic makeup, and a powerful confident expression. She stands in a dominant pose with one hand on her hip, full body visible, legs slightly apart, creating a strong runway presence. Shot from a low-angle wide lens perspective, cinematic fashion photography style. Dark studio environment with a curved backdrop, neon cyan, magenta, purple, and orange lighting reflecting across the glossy clothing surfaces. Strong rim lighting, glowing edges, dramatic shadows, futuristic atmosphere. Cyberpunk runway fashion campaign, futuristic nightlife aesthetic, sci-fi couture, luxury editorial photography, ultra realistic, realistic skin texture, detailed fabric reflections, high contrast cinematic lighting, 8K fashion photo, professional studio photography, Vogue-style futuristic editorial.
Prompt:
The image is a FRAGGY body perfume advertisement, featuring a soft portrait under cool green glass window light. On the upper left, there is a large white serif headline "FRAGGY BEYOND ORDINARY", with a line of small English text reading "Premium Fragrance Care" underneath. The upper right corner reads "FRAGGY TOTAL CARE EAU DE PARFUM". The main subject is a young woman leaning on a light-colored countertop, her long hair slightly damp, with her face resting on her arms looking into the camera. The background features vertical window frames and soft-focus light and shadow. Standing in the foreground on the bottom left is a sleek white perfume bottle, with the bottle body reading "FRAGGY", "TOTAL CARE EAU DE PARFUM", "PARIS", "FRESH MINT & AMBER", "50ml / 1.7 fl. oz", and a few delicate mist droplets on the countertop next to it. Along the left edge, vertical English text reads "FRAGGY, FOR THE MOMENTS THAT MATTER.". The bottom features large text reading "Define Your Moment", with the English translation "Define your radiance in every moment" directly below it, followed by "# Everyday Confidence" and "FROM PARIS". The overall aesthetic is clean, serene, cool-toned, and premium.
Prompt:
A futuristic luxury fashion editorial photograph of a female model wearing a pastel pink couture bodysuit with a dramatic heart-shaped chest cutout, oversized fluffy pale pink faux fur coat, and matching futuristic platform thigh-high boots. The model has sleek platinum blonde straight hair, flawless makeup, and a confident runway expression. The oversized fur coat spreads outward creating a powerful sculptural silhouette, soft fluffy texture contrasting with the tight smooth bodysuit. High-fashion cyber couture aesthetic, feminine futuristic styling, elegant and dramatic. Full body portrait, centered composition, low-angle camera perspective, model standing in a powerful fashion pose inside a dark studio. Black background with subtle gray floor reflection. Cinematic spotlight lighting from above, soft highlights on fur texture, dramatic shadows, glowing edges around hair and clothing. Luxury fashion magazine photography, futuristic runway campaign, avant-garde couture, ultra realistic, highly detailed fabric textures, realistic skin, 8K photography, cinematic depth, professional studio shoot, Vogue editorial style.
Prompt:
A futuristic high-fashion editorial photograph of a female model wearing an avant-garde metallic silver cyberpunk couture bodysuit, standing in a powerful dominant pose. The outfit features sharp architectural shoulder pieces, geometric cutouts, sculpted armor-like panels, reflective chrome textures, black futuristic fabric sections, and extreme platform boots. The model has sleek straight black hair pulled back, a confident expression, and a statuesque silhouette. Shot from a dramatic low-angle perspective with a wide-angle lens, full body visible, centered composition, long legs emphasized, powerful runway stance. Minimal futuristic studio environment with a soft gray gradient background and subtle reflective floor. Cinematic fashion lighting, soft diffused highlights, deep shadows, metallic reflections, luxury magazine photography style, futuristic runway aesthetic, high-end couture campaign, ultra realistic, sharp details, 8K fashion photography, editorial Vogue style, clean composition, realistic skin texture, dramatic atmosphere.
Krea 2 represents a useful shift in open source image generation. Instead of treating customization and usability as competing goals, it separates them into two connected stages. The Raw model gives creators room to experiment and build specialized capabilities, while Turbo focuses on delivering fast and polished results.
This approach feels closer to how real creative workflows operate experiment first, then move into efficient production. As more open-source image models adopt similar strategies, the future may not just be about having bigger models, but about creating better ecosystems around them.
 High Quality Image Gen){kind=link}