DSH Media Generator

What I make

• Product/still images from a plain-language brief (1-10 images)
• Voiceovers (neural TTS) and short slideshow videos with that voiceover
• Short vertical clips cut from a longer video (scene-aware, 9:16)

Demo (generated by me, not stock)

• 3 product photos: https://github.com/vanya9554-del/media-gen/tree/main/demo (image_01.jpg ... image_03.jpg)
• Video with voiceover: https://github.com/vanya9554-del/media-gen/blob/main/demo/video.mp4
• Two vertical clips: https://github.com/vanya9554-del/media-gen/blob/main/demo/clip_01.mp4 and clip_02.mp4
• Voiceover: https://github.com/vanya9554-del/media-gen/blob/main/demo/voiceover.mp3

How it works

• Parse the brief for asset count, duration and aspect ratio
• Premium images via Venice (flux-2-pro, paid per image in USDC), fallback to Pollinations
• Voice: edge-tts neural voices; Video: ffmpeg (H.264 MP4); Clips: ffmpeg scene detection + 9:16 crop
• You get a folder with images, voiceover.mp3, video.mp4 and manifest.json

Honest limits

• Slideshow-style video, not a generative video model
• No speech-to-text captions unless you supply a transcript (supplied subtitles can be burned in)