Yeri AI image-to-video experiments: prompts, models and edits

Yeri AI bundles text-to-image, image-to-image, text-to-video and image-to-video generation into one web app, and this repo collects what I learned using it. A plain Markdown file is the easiest way I have found to keep prompt variants, model choices and settings together.

Video

I generate a clean still first, then animate it. That gives much more control than describing everything in one video prompt. Text-to-video clips are short, so plan one idea per clip and cut several together afterwards.

Tips

• Keep a log of prompt, model and seed for every image you plan to reuse.
• Check hands, text and logos at full size before publishing.
• Use the same reference image across a series to keep characters consistent.

Use cases

Marketers, e-commerce sellers, designers and video creators are the people who get the most out of it. I mostly use it for blog headers and short social videos, but it also works for quick mockups when pitching an idea.

Image prompts

Start with the subject, then the setting, then light and lens. For example: 'ceramic mug on a linen table, soft window light from the left, 50mm, shallow depth of field'. When matching a brand look, upload one or two reference images and describe only what should change; the model keeps the rest.

Templates

Video templates and photo effects are the fastest way to get something social-ready; swap in your own photo and adjust the text. The prompt library is useful when you are stuck: copy a prompt, change the subject, keep the lighting and composition terms.

Models

Model choice matters more than prompt length. Fast models are fine for exploring ideas; switch to a pro model only for the final render to save credits. Yeri AI lets you choose the model per job. Seedream is my pick for detailed product shots, Nano Banana for fast drafts and multi-reference edits, and GPT Image when the picture has to contain readable text.

Links

Official site: https://yeri.ai

Corrections and new prompt ideas are welcome.