Since a bit more than a month I have tried the website PixelDojo and their WAN 2,2 AI model:
This works really great!!!
Cost is about $1.50 per minute. It allows to use the end frame of a video to be the starting frame of the next video - allowing quite long video sequences. Quality is excellent, rarely there are any artifacts.
And if provided explicitly with start and end pic, one really can tell a great visual story.
Synexa provides access to a variety of tools: WAN 2.1 (start and end image), Framepack (single image) for video. Also several image generation tools (e.g. Hunyuan, flux, stable-diffusion) are available, and audio generation for videos. The cost is reasonable: begins at $0.10 per video (5 seconds).
NSFW content is not objected, it is possible to upload NSFW images as basis for the video generation. The result is, however, not always convincing: there can be weird artifacts in the Framepack model, and the WAN model where one can specify a start and an end image, there is not necessary the end image of the video the one that was originally provided. The VIDU Q1 at Civitai works here much more reliable. Also there are no LORAs available, which means one cannot extend the model with customised behaviours.
But considering the low cost, one can experiment a bit and get reasonable results.
