easyai.video
What things cost, what breaks, and what we can actually measure. Prices and findings carry the date they were taken.
Featured collection
The AI-video market changes every week. We are tracking it: dated prices now, comparable quality tests as they are run.
Quality scoring is not published yet. A rubric has to be defined and applied to more than one test round before any ranking would mean anything.
ComparisonsDated prices and like-for-like tests.
GuidesHow to direct a shot and evaluate the result.
Product notesWorkflow decisions, measurements and changes.
Recasting people, locations and colours in a finished clip: four replacement routes judged on whether the people still move, why Wan 3.0 needs short runs, and why outpainting 9:16 to 16:9 shows.
One miniature comedy scene shows how model choice and specialised prompt skills affect picture consistency, dialogue, camera moves and production costs.
One selfie clip, seven ways to change the room and the clothes around the speaker: restyle, cut-out and photo-avatar routes, with the process, prompts and a reusable template.
Reference sheets for every recurring element, how to get them into Seedance, and why fewer references hold a story together better.
Topaz AI and Astra settings, plus real Higgsfield UI and ByteDance Pro upscales with before-and-after frames and recorded settings.
Seedance 2.5 Editor, Beeble SwitchX, Veo Omni Flash, Higgsfield Genjutsu, Runway Aleph 2.0 and Kling O1 Edit for changing one thing in a finished clip, with a real Genjutsu job and the objects it missed.
Six running-commercial takes show how the reference frame, prompt direction and model choice combine to shape the final ad.
A published “flat” per-second rate excluded the resolution we actually wanted. Reconciliation against one paid run exposed the 4.4× gap.
Sixteen service and model combinations reduced to a per-second rate, with the resolution, pricing basis and limits that make the headline numbers unlike each other.