Unfinished clip
Before you make anything
A short lesson about copyright, trademarks and people's likeness, then one agreement. About two minutes. You can't make pictures or clips until you agree.
Get the video model
Getting ready
Your phone is warm, so it's slowing down to stay cool. This one will take a little longer.
Keep p0pc0rn open — the screen stays on.
Details
Check the frames
What this means
Shape
Quality
Model
Photo detail
Size
Slow, may run out of memory. About 15 minutes on an iPhone; 832×480 is the safe choice.
Quality
For high quality renderings use 9B with first frame or first frame last frame.
Length
More
Seed
Each picture takes 2–3 minutes. Plug in, and keep p0pc0rn open until it's done.
Get the assistant
Before the assistant makes anything
A short lesson about copyright, trademarks and people's likeness, then one agreement. About two minutes.
Ask for a song, a picture or a clip
The assistant writes the prompt, the song's style and its lyrics for you, then the job waits its turn in the queue. Everything runs on this phone.
Two pictures make a clip between them.
This browser can't keep pictures between visits. Save the ones you like to Photos.
Pictures, clips and songs you make stay on this phone.
Models
Android (Chrome): experimental. Prepared on a desktop with Android-like limits; not yet tested on a phone. Run Settings › About › Device check first and send Josh the report.
Deleting is not available while something is being made.
Cast
People you can name in a request: “make Sam say …” speaks in their own recorded voice, and their photos put them in a picture, a clip or a music video. Each one is added with their permission and stays on this phone.
Download Voice in Models above to make anyone here speak.
Video decode memory
How much graphics memory a clip may use while it develops. 700 MB is the fastest; if a clip stops while developing, choose 450 MB.
Assistant
Assistant memory
Storage
On this phone
Checking…
Removes only files that neither model nor the Library uses. You'll see the list first.
Rights & Copyright
The lesson
If someone copies your work
Reaching AI and cloud providers
AI cases to know
Heat
A picture takes 2–3 minutes, a 5-second clip about 6 and a song a minute or two. After about a minute the phone warms and slows down. Plugged in, out of the sun, case off, and a short break between jobs all help.
About
p0pc0rn
by Joshua Hughes
Diagnostics
The last 200 steps on this phone: what ran, its memory and any errors. Copy adds the app version, the browser's own description and, where the browser reports it, the phone's memory. Research and Cast steps keep kinds and counts, never a topic, title, quote or name. Nothing leaves the phone unless you copy it.
FLUX.2 [klein] 4B and FLUX.2 autoencoder © Black Forest Labs, Apache-2.0 · Qwen3-4B text encoder © 2024 Alibaba Cloud, Apache-2.0 · SANA-Video 2.0 5B © NVIDIA, Apache-2.0 · LTX-2.3 video decoder and picture reader (VAE decoder and encoder) © Lightricks, LTX-2 Community License Agreement · tokenizers.js © Hugging Face, Apache-2.0. The model files are modified (re-quantised) copies. The popcorn progress animation was generated with LTX-2.3 (Lightricks) from FLUX.2 [klein] stills.
The assistant: Gemma 4 E2B-it © Google, Apache-2.0 (not the Gemma Terms of Use). Songs: ACE-Step 1.5 turbo, its 5 Hz lyrics planner and its audio decoder © ACE Studio and StepFun, MIT; the Qwen3-Embedding-0.6B text encoder inside it © Alibaba Cloud, Apache-2.0. These model files are modified (re-quantised) copies too. Lookups use Wikipedia (CC BY-SA 4.0) and Fandom wikis (CC BY-SA, as each wiki states): the assistant cites the pages it used.
Cast voices: Qwen3-TTS-12Hz-0.6B-Base and its 12 Hz codec © Alibaba Cloud (Qwen), Apache-2.0 (re-quantised copies). Face finding for Cast photos: MediaPipe BlazeFace © Google, Apache-2.0 (engine/face/LICENSE-mediapipe.txt). Lip-sync (experimental): MuseTalk 1.5 UNet (MIT; experimental, for consented faces and voices only; its CreativeML Open RAIL-M use restrictions are part of your terms), sd-vae-ft-mse © Stability AI (MIT), whisper-tiny encoder © OpenAI (MIT), MediaPipe Face Landmarker © Google (Apache-2.0), re-quantised copies.
Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms. The video model reads your words with Gemma-2-2B-it; using it is subject to the Gemma Prohibited Use Policy, and the use restrictions of the LTX-2 Community License (paragraph 4 and Attachment A). Both are part of p0pc0rn's terms of use.
Model licencesRoad map
🖼️ FLUX 3 [klein] image model as the Better picture option.
🔊 Sound for clips: sound effects made with your videos.
✈️ Complete airplane mode (coming soon): once p0pc0rn is on your Home Screen it runs fully offline, and updates only when you choose.
👀 Live Preview of renders: watch your picture or video take shape while it renders.
👄 LoRAs for SANA video, starting with a lip-dub LoRA like LTX's.
🎬 Animated cartoons, sitcoms and cinema, rendered locally and free on your own device.
🎷 Labs: experimental features, starting with Playing Jazz — give it one idea and the AI improvises a short film, each clip picking up from the last frame of the one before, scored with an instrumental track.
🔄 Updated workflows.
🐞 Debugging.
✨ More features.
The app is updated periodically, when Josh Hughes decides to.
Connect