🌀 LoRA – Logo Animation via Image-to-Video | Experimental ReleaseAfter a series of experiments training this LoRA focused on animating logos from static images, I’ve reached a version that’s usable and worth sharing. This is not the final version — I plan to improve quality further soon, but this is already a solid base. This model was trained on Wan I2V 14B 720P, and uses the trigger word: <code>4n1m4t3dl0g0</code>. Sometimes some results may be degraded, try generating again with another seed or change the prompt by adding more details. 🔧 Training Setup<ul><li>Base model: Wan I2V 14B 720P</li><li>Trained using: <a target="_new" rel="ugc" href="https://github.com/alisson-anjos/diffusion-pipe-ui/tree/new-ui">Diffusion Pipe UI (custom fork)</a></li></ul>Future versions will also test Wan I2V 14B 480P, and Wan T2V (1.3B and 14B).🎞️ Dataset<ul><li>170 hand-selected video clips</li><li>Resolution: 244p</li><li>Duration: 33, 49 frames average</li><li>Over a week of curation and prep </li></ul>🧠 Captions via LLM Video captions were generated using <a target="_new" rel="ugc" href="https://ollama.com/huihui_ai/gemma3-abliterated/tags">Gemma3 Obliterated 27B (30GB)</a> — an uncensored variant of Gemma suitable for NSFW content. I used this via the “Ollama Describer” custom node inside ComfyUI, enabling automatic caption generation directly from the videos. 🪄 Prompt Template (for generating custom captions)To help others generate their own logo animation prompt, here’s a ready-to-use prompt template for ChatGPT, Ollama, or any LLM: <pre><code>You are an expert in creating short animation-style captions for logo videos.
Your task is to analyze the provided image and generate a single-paragraph prompt, under 120 tokens, describing how this logo could be animated in a smooth and visually pleasing way — suitable for image-to-video models.

The animation must preserve the visual identity of the main character or logo exactly as shown in the image. You may describe enhancements like lighting effects, text appearance, soft movements, or ambient motion, but do not alter the character's shape, style, or features in any way. No major changes or reimagining of the subject are allowed.

Use this format as inspiration, phrased naturally and clearly:

4n1m4t3dl0g0, [main visual element from the image], [its subtle or clean animation]. The word “[TEXT]” [how it appears], with [gentle or stylish letter effects]. The whole logo [global animation effect]. Background is [a simplified or matching version of the image background].

Use natural, flowing phrasing. Do not list items. Avoid complex or abstract ideas that could deform the visual. Keep character and text grounded, and use soft animation styles that highlight — not distort — what’s already present in the image.

Examples:

4n1m4t3dl0g0, no visual element besides text. The word “ChamaLeOn” forms in a wave motion, each letter scaling up and bouncing into place. A neon gradient flows across the word. The whole logo softly pulses. Background is solid black.

4n1m4t3dl0g0, a cartoon mouse with goggles launches upward using a small red rocket on its back. A puff of smoke appears under it. The words “ROCKET MOUSE” fade in from the right, glowing slightly. The entire logo bounces gently in place. Background is flat white with a blue circle.

You must always return a single paragraph. Do not provide commentary, explanation, or more than one version. Keep the character visually consistent with the input image.</code></pre><hr /><h3 id="if-you-speak-portuguese-join-our-community-on-discord!-share-your-results-and-get-buzz!-mf4207ti4"> If you speak Portuguese, join our community on <a target="_blank" rel="ugc" href="https://discord.gg/NcxPfbYedp">discord</a>! 📢 Share Your Results and Get Buzz!</h3>If you use this LoRA to animate your logo, I'd love to see it!

Files with SHA256: d91527165dee8a35417a912b8c29e4fa8c64e809e72d4a47258b8922f5fdaf4e