AI BABES — Studio

👛

AI BABES

AI Fashion Video Studio · try-on images · motion videos · one click to post-ready

v53

📜 What's new

v53 — every generated video now has its own ⬆️ upscale link (uses the factor picked in the Upscale card). v52 — ⬆️ Upscale ALL button: one click upscales every generated video (by URL, no re-upload). v51 — ⬆️ Upscale card (Topaz 2×/4×): drop any clip, get a sharper/higher-res copy. v50 — FIX: a load-order bug (v47 motion library) was breaking Quick animate + all buttons on some loads; resolved. v49 — Kling 3.0 Motion Control 1080p pro added (~$0.18/s est, max quality). v47 (🎬 MOTION LIBRARY: drop up to 30 motion clips ONCE — each gets a number; ▶ use sets the active motion for single/Quick-animate runs, batch pair rows pick a # from a dropdown; stored in localStorage, upload links expire ~3 days then marked ⚠️ stale for re-drop. v46: 🎭 clips SKIP the 🤳 shake automatically — motion control already carries the real camera from the ref; shake now only applies to seedance/kling i2v clips. v45: 💰 motion-control cost estimate uses the ⏱️ length dropdown until a motion vid is dropped — no more scary 30s worst-case guess. v44: 🎭×N BATCH PAIRS: add pairs of picture + motion video, generate all in parallel — 5 pics × 5 different motions in one click; shared model/framing settings; results save/ZIP/shake like everything else. v43: 🎭 selecting a motion-control model HIDES the Motion style dropdown + video prompt — a note points to the 🎭 card instead; prompts reappear when you switch back. v42: 🎭 sends NO prompt by default — pure motion copy; only text typed in the 🎭 scene box is sent, motion-preset prompts are ignored for 🎭 models. v41: 🧭 orientation toggle — framing follows your PICTURE (max 10s, best fidelity) or the MOTION VIDEO (up to 30s); cost readout follows the cap. v40: Kling 2.6 — ~$0.07/s 720p / ~$0.14/s 1080p — and Kling 3.0 — ~$0.09/s, best identity — in the Video model dropdown; drop your girl + a motion video in the 🎭 card, works from the card AND from product rows / Quick animate; output = motion vid length, max 10s; prices are fal/WaveSpeed estimates — VERIFY vs credits after run #1. Plus v39's 🤳 natural shake on every save)

🔑 API Key

Saved in this browser only — never sent anywhere except the generation API.

🎞️ Video model

📐 Aspect ratio

⏱️ Video length

Kling: 5/10 only. Grok: always ~6s.

🖼️ Image model

🔍 Image resolution

Applies to both image models now.

🎬 The look — pick these two first

📸 Setting — where the photo happens

✏️ Fills the try-on image prompt.

🎥 Motion style — what she does in the video

✏️ Fills the video motion prompt.

💡 Match the pair — 🚗 setting with a 🚗 motion, ✋ setting with ✋ motion. Full prompt text lives in ✏️ Edit master prompts down in Step 2; your edits there save automatically.

👤 Step 1 — Your girl (reference image)

The locked mirror-selfie reference. Uploaded to temp storage on first generate (links live ~3 days, re-upload happens automatically per session).

⚡ Quick animate — drop ready pics 📥, get videos 🎬. Add as many as you want.

Drag pictures anywhere in this box, or use the button. You can add them in several batches — new ones are added to the list, not replacing it.

No pictures yet — each one you add becomes its own video, all generated at the same time.

🤳 Natural shake — real handheld camera movement, measured from the lilypads originals

Add natural shake to every video I save (⬇ save video, ZIP, one-by-one) 🎭 motion-control clips are ALWAYS skipped — their camera movement is real, from the reference video.

AI video models output a tripod-still frame no matter what the prompt says. This pass adds the comp's measured motion — slow up/down + side-to-side sway, fast hand tremor, micro-roll — with a new random path per clip, so no two posts move the same. First use downloads ffmpeg (~31MB, one time). ~30–90s per clip.

Already have videos? Drop mp4 files here (or use the button) — you'll get shaken copies back.

⬆️ Upscale (Topaz) — sharpen + boost resolution of any clip

🔍 Upscale factor

Runs server-side (Topaz). Output length = input length. Prices are estimates — check the credits readout after run #1.

Drop mp4 files here (or use the button) — you'll get upscaled copies back. Uploads over ~9MB auto-compress before sending.

🎭 Motion control — drop your girl + a motion video, she copies the moves (Kling)

👩 Character picture (your ai babe)

🕺 Motion video (3–30s mp4 — the moves to copy)

🧭 Framing follows…

Keep the motion video's background (Kling 3.0 only)

✏️ Scene prompt (optional — LEAVE EMPTY to just copy the motion, recommended)

The model is picked in 🎞️ Video model above (the 🎭 entries). Works from here with the picture on the left, AND from product rows / ⚡ Quick animate when a 🎭 model is selected — every try-on pic gets the dropped motion. 🧭 "your PICTURE" = best outfit/identity fidelity, max 10s; "the MOTION VIDEO" = up to 30s but framing comes from the ref video. Uploads over ~9MB auto-compress. Prices are fal/WaveSpeed estimates — verify against the credits readout after run #1. Saved videos go through 🤳 natural shake automatically if it's on.

🎬 Motion library — your numbered moves (pick by #, no file hunting)

Drop your best motion clips once (up to 30) — each gets a number, in order. They upload to secure storage right away; links live ~3 days, stale ones show ⚠️ and just need a re-drop. Click ▶ to make a number the ACTIVE motion (used by the single generate above and by ⚡ Quick animate). Batch pair rows can pick a # directly.

🎭×N Batch pairs — each picture with its OWN motion video, all generated at the same time

Uses the same 🎞️ model / 🧭 framing / background settings above for every pair. Want 5 pics with the SAME motion instead? Drop ONE motion video above and use ⚡ Quick animate.

🛍️ Step 2 — Products

✏️ Edit master prompts (image + video) — only if you want to reword something

🖼️ Try-on image prompt (garment description gets appended) 🎥 Video motion prompt

The woman is taking a mirror selfie with the phone covering her face the entire time. She has exactly two arms and two hands and no more. One hand holds the phone up in front of her face for the entire video and never lets go of it and never leaves the frame. Her other hand is the ONLY free hand in the video: it starts exactly where it is in the first frame and moves in one single continuous path, never teleporting or reappearing somewhere else. A third hand, a third arm, or a stray hand belonging to nobody must NEVER appear anywhere in the frame, including in the mirror reflection and at the edges of the frame. Her body, her reflection and the room contain no other people and no other hands. One hand holds the phone up the whole video. Sequence: she starts on a full-body view, gently swaying, her free hand relaxed at her side. She then takes two slow steps toward the mirror so the framing tightens to a head-to-hips view. Up close, her free hand touches the outfit in smooth continuous motions: patting and smoothing the fabric over her stomach, lightly tugging the hem, brushing over the material to show its texture. Then she places her free hand on her hip and turns to the side to show the side profile of the outfit, glancing down at it. Finally she steps back to the full-body view and settles, still swaying. The camera is handheld and moves with her the whole video: rising and dipping with her steps, drifting, wobbling and micro-tilting, never still, like a phone held in one hand while walking. No zoom, no cuts, no fast whips. Warm lamp lighting stays constant. Realistic amateur smartphone mirror selfie video, 9:16.

🔧 Debug log — screenshot this if anything breaks