<h3 id="ultrasharpcc-viral-style-sharpness-and-color-correction-lora-882pg0gt6">🧬 UltraSharpCC – Viral-Style Sharpness &amp; Color Correction LoRA</h3>Model: Wan T2V 14B Compatibility: VACE (Kijai Version) – <code>Image2Video</code>, <code>First Frame to Video</code>, <code>Mask to Video (</code><a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan2_1-T2V-14B_fp8_e4m3fn.safetensors">Wan2_1-T2V-14B_fp8_e4m3fn.safetensors · Kijai/WanVideo_comfy at main</a><code>) + (</code><a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan2_1-VACE_module_14B_fp8_e4m3fn.safetensors">Wan2_1-VACE_module_14B_fp8_e4m3fn.safetensors · Kijai/WanVideo_comfy at main)</a> If you want to use it with I2V you can use the Wan T2V + the VACE module extracted by Kijai.<a rel="ugc" href="https://pastebin.com/5n8gu1XD">Workflow Wan T2V 14b + VACE Module </a> Optimized for: Use with CausVid (8–10 steps fast generation)(<a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan21_CausVid_14B_T2V_lora_rank32_v2.safetensors">Wan21_CausVid_14B_T2V_lora_rank32_v2.safetensors · Kijai/WanVideo_comfy at main)</a>or Lightx2v (4 - 10 steps fast generation using LCM) (<a target="_blank" rel="ugc" href="https://huggingface.co/Kijai/WanVideo_comfy/blob/main/Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors">Wan21_T2V_14B_lightx2v_cfg_step_distill_lora_rank32.safetensors · Kijai/WanVideo_comfy at main)</a><hr />UltraSharpCC is a visual enhancement LoRA designed for video generation using Wan T2V 14B. It simulates the viral video look popularized on TikTok, where perceived image quality is boosted through sharpness, high dynamic range glow, and bold color grading—often reminiscent of Topaz filters and fake-4K aesthetics.This LoRA enhances clarity, contrast, and surface detail without altering the original artistic style, making it ideal for transforming regular image or video prompts into cinematic clips that look dramatically upscale.It is fully compatible with the VACE system, especially in the following modes:<ul><li><code>Image2Video</code></li><li><code>First Frame to Video</code></li><li><code>Mask to Video</code></li></ul>UltraSharpCC also works seamlessly with CausVid, enabling ultra-fast video generation in just 8 to 10 steps with minimal quality loss, making it perfect for workflows that prioritize speed and efficiency.<hr /><h3 id="training-details:-981w9du3f">🧪 Training Details:</h3>V1<ul><li>Framework: <a target="_blank" rel="ugc" href="https://github.com/tdrussell/diffusion-pipe">Diffusion Pipe</a></li><li>Epochs: 26</li><li>Batch Size: 1</li><li>Rank: 64</li><li>Optimizer: automagic</li><li>Resolution: – Videos at 512px – Images at 1024px</li><li>Dataset: – 99 short videos – 100 high-resolution images</li><li>Captions: Generated using a custom LLM (gemma3:12b) prompt focused on visual quality (see below).</li></ul>V2<ul><li>Framework: <a target="_blank" rel="ugc" href="https://github.com/tdrussell/diffusion-pipe">Diffusion Pipe</a></li><li>Epochs: 76</li><li>Batch Size: 4</li><li>Rank: 64</li><li>Optimizer: automagic</li><li>Resolution: – Videos at [512, 288]</li><li>Dataset: – 99 short videos</li><li>Captions: Generated using a custom LLM (gemma3:12b) prompt focused on visual quality (see below).</li></ul><hr /><h3 id="prompt-template-used-for-captions-(llm-friendly):-6gkq12h96">💬 Prompt Template Used for Captions (LLM-friendly):</h3>Analyze the content of this video frame sequence and return a single-paragraph description that includes the following: sh4rpn3ss followed by a detailed explanation of the visual quality enhancements applied to the video (e.g., increased sharpness, 4k, 8k, HDR glow, crisp outlines), and a focused description of the main character (if present), including their appearance and the visual style of the video (e.g., anime, cartoon, CGI, live-action). The description must be concise and capture both the enhancement effects and the artistic style. Do not include any formatting, metadata, or comments—only output a single paragraph starting with sh4rpn3ss.You can use this prompt with any LLM (like Gemini, GPT, Mistral, or Qwen) to generate captions for your own dataset or to describe generated videos in a consistent, quality-focused format.<hr /><h3 id="usage-tips:-zr2o2f0e5">✅ Usage Tips:</h3><ul><li>Add the trigger word <code>sh4rpn3ss</code> to your prompt to activate the LoRA’s effects</li><li>Works best with portraits, stylized characters, and cinematic lighting</li><li>Ideal for viral short videos, AI-generated live wallpapers, and motion-enhanced artworks</li><li>Combine with CausVid for high-speed rendering (8–10 steps) with impressive visual fidelity</li></ul>

Files with SHA256: 7c5ae72b45f021bacff362ba55fb5f58935df2b24258436bdf1d8fb461ee71b2