The latest articles on DEV Community by Nana (@nana_banana). https://dev.to/nana_banana https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4070082%2F6accfc34-41d7-4e72-9d76-2228aeb97754.png https://dev.to/nana_banana en Nana Tue, 18 Aug 2026 11:00:04 +0000 https://dev.to/nana_banana/h3-video-creating-ai-videos-from-text-and-images-with-minimax-h3-3gc9 https://dev.to/nana_banana/h3-video-creating-ai-videos-from-text-and-images-with-minimax-h3-3gc9 <h1> H3 Video: AI video generation in the browser </h1> <p>H3 Video is a browser-based AI video generator powered by MiniMax H3. It helps creators and small teams turn a written idea or an image into a polished video without installing desktop editing software.</p> <h2> What you can create </h2> <p>You can start with a text prompt for text-to-video generation, or use image-to-video workflows. H3 Video supports first and last frames, reference images, sound, flexible aspect ratios, and saving or downloading finished creations. These controls make it easier to guide motion, preserve a visual style, and iterate quickly.</p> <p>Try H3 Video here: <a href="https://h3video.io/" rel="noopener noreferrer">https://h3video.io/</a></p> <h2> Who it is for </h2> <p>The tool is useful for social creators, marketers, product teams, and anyone who needs short visual clips from an idea or an image. The workflow runs in the browser, so it is easy to experiment with prompts and compare results.</p> <p>This post is an independent overview of the product and its current capabilities.</p> ai Nana Mon, 10 Aug 2026 15:41:11 +0000 https://dev.to/nana_banana/seedance-25-ai-video-generator-prompting-notes-from-a-week-of-testing-28h1 https://dev.to/nana_banana/seedance-25-ai-video-generator-prompting-notes-from-a-week-of-testing-28h1 <p>I spent a week stress-testing <strong>Seedance 2.5 AI Video Generator</strong> — text-to-video, image-to-video, long clips, weird prompts, the whole range. These are the prompting notes that actually made a difference.</p> <h2> 1. Be concrete about camera movement </h2> <p>Vague motion words produce vague results. Instead of "a drone shot", try "slow aerial push-in from 200m to 30m over a pine forest, slight parallax, golden hour". The model is surprisingly good at honoring explicit camera vocabulary — it falls apart when you leave motion implicit.</p> <h2> 2. Use image references for consistency </h2> <p>Seedance 2.5 accepts up to 50 multimodal references. The single biggest quality jump came from feeding it a character sheet or a style frame and letting the clip inherit it. Text-only consistency across multiple shots is still unreliable; reference images fix most of that.</p> <h2> 3. 30 seconds is a real constraint </h2> <p>The 30-second clip limit sounds generous until you try to tell a story. What works: treat each clip as one continuous shot or one idea, not a mini-movie. Plan cuts in post, not in the prompt.</p> <h2> 4. Native 4K is not free </h2> <p>Renders are genuinely 4K (no upscale pass), but longer clips at 4K take a while and the preview is lower-res. Prompt, iterate on previews, and only render final takes at full resolution.</p> <h2> 5. Keep a failure log </h2> <p>My best results came after logging what broke: multi-subject scenes drift, text in frames is still unreliable, and extreme close-ups on faces occasionally warp. Knowing the failure modes beats blind re-rolling.</p> <h2> Where to try it </h2> <p>If you want to test the same workflows, the tool is at <a href="https://seedance2-5.me" rel="noopener noreferrer">Seedance 2.5 AI Video Generator</a>. Free to try, and the reference-image workflow is where it shines.</p> <p><em>Full disclosure: I'm part of the team behind this tool. This post was drafted with AI assistance and edited by me. All prompts and results above are from my own testing.</em></p> ai showdev webdev