1. Workflow Overview

This workflow, named “Ultra-HD Cinematic Photography”, generates high-resolution images with film-grade quality, suitable for posters, concept art, or photo enhancement. Key features:

• 4K Output: Supports custom resolutions (e.g., 1536x768).

• Multi-Stage Optimization: Combines LoRAs, upscaling, and post-processing filters.

• Cinematic Control: Fine-tunes lighting, color, and composition via text prompts.

2. Core Models

1. Base Model:

   • F.1基础模型_fp16: General-purpose realistic generation.

2. LoRAs:

   • 鲜创一派@光域史诗_超高清电影级摄影_F.1: Enhances cinematic lighting.

   • 鲜创一派@细节狂魔_细节增多增强呈现更多细节_F.1: Boosts texture details.

3. Upscaler:

   • 2xNomosUni_span_multijpg_ldl: 2x super-resolution.

4. Depth Estimation:

   • depth_anything_v2_vitl.pth: Generates depth maps for post-processing.

3. Key Nodes

• Text Encoding:

  • CLIPTextEncodeFlux: Processes complex prompts (e.g., "dramatic chiaroscuro").

  • BaiduTranslateNode: Auto-translates Chinese/English prompts.

• Generation:

  • KSampler: 30 steps, Euler sampler, CFG=1.

  • CR Aspect Ratio: Sets aspect ratio (e.g., 1536x768).

• Post-Processing:

  • UltimateSDUpscale: Tile-based upscaling to 4K.

  • LayerFilter: FilmV2: Adds film grain, lens flares.

  • DepthAnythingV2Preprocessor: Depth map for DoF effects.

4. Workflow Structure (Groups)

1. Text-to-Image

   • Input: Cinematic prompts (e.g., "movie poster style, dark gold palette").

   • Output: Base generated image.

2. Control Panel (Model Loading)

   • Loads base model, LoRAs, and VAE (ae.sft).

3. Upscaling

   • Uses UltimateSDUpscale with LoRA-enhanced details.

4. Post-Processing

   • Applies FilmV2 for color grading and effects.

5. Inputs & Outputs

• Inputs:

  • Positive prompt (e.g., "hyper-detailed eye close-up, wet skin texture").

  • Resolution (default 1536x768).

  • LoRA weights (e.g., 光域史诗=1.0).

• Output:

  • 4K image with side-by-side comparison (Image Comparer).

6. Notes

1. Dependencies:

   • Install:

     • ComfyUI-UltimateSDUpscale (tiled upscaling).

     • DepthAnything (depth maps).

     • LayerFilter series (post-effects).

2. Hardware:

   • VRAM ≥12GB (4K upscaling is resource-intensive).

3. Optimization:

   • Reduce UltimateSDUpscale tile size (e.g., 32x32) to avoid OOM.

   • Use fp8_e4m3fn precision for lower VRAM usage.