CIVITAI / Workflows

Krea2 8-Step Cinematic Portrait Benchmark Workflow

Watch the full video first if you want to understand how this Krea2 8-step cinematic portrait benchmark workflow works in practice. The video shows how a lightweight Krea2 Turbo pipeline can generate polished portrait-style images with a simple structure, fast sampling, and a tuned conditioning rebalance setup. This ComfyUI workflow is designed as a clean Krea2 Turbo benchmark for cinematic portrait and stylized image generation. Its purpose is not to be a large multi-stage production graph, but to provide a fast, stable, and repeatable baseline that creators can use to test Krea2 image quality, prompt response, composition, lighting, and cinematic style control. The workflow is built around krea2_turbo_bf16.safetensors as the main image generation model. The text encoder route uses qwen3vl_4b_fp8_scaled.safetensors with the Krea2 CLIP type, giving the workflow a compact but efficient text-conditioning setup. The VAE route uses qwen_image_vae.safetensors for final image decoding. This makes the graph lightweight and direct, suitable for quick testing and RunningHub online use. The key control module in this workflow is ConditioningKrea2Rebalance. The workflow uses a custom per-layer weight setup with a multiplier of 2.5. This rebalance node is used to strengthen the positive conditioning and improve the model’s response to cinematic prompt structure, visual hierarchy, subject placement, lighting, and style description. It is one of the main reasons this workflow can feel more controlled than a plain Krea2 Turbo render. The sampling section is intentionally simple. It uses KSampler with 8 steps, CFG 1, euler sampler, simple scheduler, and full denoise. This makes the workflow fast enough for practical iteration while still keeping enough quality for cinematic visual tests. The fixed seed also makes it easier to compare prompts, rebalance settings, model behavior, and resolution changes across different runs. The resolution section uses FluxResolutionNode to provide flexible aspect-ratio and size control. In the uploaded setup, the workflow is prepared for high-impact composition testing, including ultra-wide cinematic framing. This is useful for creators who want to test poster-like portraits, fantasy character shots, cinematic covers, key visuals, thumbnail art, and stylized AI image concepts. The negative conditioning route uses ConditioningZeroOut, k