📼 ZIT STYLER: SEEDVR VHS & CRT WORKFLOW

SquiDavey's ZIT Workflow combines the speed of Z-Image Turbo with a local LLM to generate improve your prompts, upscales with SeedVR, then processes the image through CRT Post Process Suite.

🛠️ HOW IT WORKS

1. LLM Expansion: A local LLM (via LM Studio) expands simple concepts into rich, photorealistic descriptions.

2. Generation: Z-Image Turbo creates the base image efficiently.

3. Analog FX: The image passes through a CRT/VHS simulation for authentic glitch and signal noise.

4. Upscale: The SeedVR DiT model performs a tiled upscale on the final result.

📥 REQUIRED MODELS & DOWNLOADS

1. Z-Image Turbo (The Generator)

Place these in your ComfyUI/models/ directory:

• UNET: z_image_turbo_bf16.safetensors
  Folder: models/diffusion_models/

• VAE: ae.safetensors
  Folder: models/vae/

• CLIP: qwen_3_4b.safetensors
  Folder: models/text_encoders/

2. SeedVR (The Upscaler)

• DiT Model: seedvr2_ema_7b_fp16.safetensors
  Folder: models/checkpoints/

• VAE: ema_vae_fp16.safetensors
  Folder: models/vae/

3. Post-Process (CRT Effect)

• Model: 4x-ClearRealityV1.pth
  Folder: models/upscale_models/

4. Local LLM (Prompting)

• Software: LM Studio
  Instructions: Download a model (e.g., Qwen 2.5 7B), go to the Developer/Server tab, and Start Server on port 1234.

⚡ WORKFLOW INSTRUCTIONS

MODE A: Text-to-Image (New Generation)

1. Ensure LM Studio server is running.

2. Type a simple concept in the Prompt group (Green Node).

3. Set the "SEEDVR / LOAD IMAGE" switch to True.

4. Press Queue Prompt.

MODE B: Image-to-Upscale (Existing Image)

1. Set the "SEEDVR / LOAD IMAGE" switch to False.

2. (Optional) Disable the "LLM" group to save time.

3. Upload your image into the "UPSCALE BYPASS" node.

4. Press Queue Prompt.

Note: If you run out of VRAM during the Upscale step, try reducing the tile size in the SeedVR settings and increase the blocks to swap.

Enjoy 🥳