A facility from the Weidu Fanling collective

Clone Chat

A short chat beside a memory.

Keep a photo or play a recording beside a conversation that helps you reflect on your feelings and memories. Replies are generated by a language model on your device. Personal questions show only the notes you supplied. The facility does not recreate the person in your recording, and its checks cannot guarantee accuracy. You can choose generated speech and portrait animation after downloading their models and confirming permission.

Run the conversation on your device

Start a small model on this device

Qwen 2.5 0.5B runs in this browser. The first download is about 297 MB. It needs a browser with WebGPU, enough storage and more than 1 GB of graphics memory. That memory figure is approximate; this page and its longer context need additional memory. Some devices cannot run it, and its replies can be inaccurate.

Model files are downloaded from public hosts and may stay in this browser's cache. Later visits can reuse them. Ordinary page and download requests still reach the hosting services. Your chat text, photo and recording are not sent to them. There are no accounts, analytics or cloud chat fallbacks here.

I want to download or reuse this model and run it on this device.

No model is loaded. Nothing will download until you choose an action below.

Unload frees the model's graphics resources. Removing files clears only this release's small Qwen model files, which may also be used by other Weidu Lab tools. It does not delete saved conversations or other website data. Removal may load the small runtime, but does not download model weights.

Runtime and model licences

Optional generated speech

Prepare a voice with permission

Generate speech from a short voice sample, separately from the conversation model. Processing stays on your device. The result is a new recording of generated words, not a recording of the speaker saying them.

Use your own voice or an adult's voice with their explicit permission for generated speech. If their wishes are unknown, keep their real recording in the memory player. That player is separate from the voice sample. No microphone permission is needed.

Whose voice is this?My own voiceAnother adult's voice, with their explicit permissionI am 18 or older. The speaker is 18 or older and has explicitly agreed to this voice sample being used to generate speech in this voice. I have the right to use the recording.I understand that generated speech is artificial. It is not a recording of the speaker saying these words, and I will not present it as one.

A clear audio sample, about 5 to 10 seconds

Agree above before choosing a voice sample. Audio stays on this device and is not included in conversation files.

Pocket TTS uses the CPU. Its first model download is about 146 MB, plus the runtime. Preparation and speech generation may take time. Model files may remain in this browser's cache; unloading releases the running model, not its cached files.

I want to download or reuse the local voice model on this device.

No voice model is loaded. Nothing is downloaded until you choose to start it.

After a text reply is finished, choose Speak this reply beside that reply. No reply is turned into speech automatically. Use the generated speech player to listen after choosing a reply. Delete and start again clears the sample, prepared voice, generated audio and the running text, voice and portrait models.

Voice software and model licences

Optional portrait animation

Animate a permitted portrait

Animate a permitted portrait with the generated speech you choose. Processing stays on your device. Movement is generated and approximate, and can look slow or uneven even on a recent computer. This is new footage, not an authentic recording of the person speaking.

The first model download is about 78 MB, plus the runtime. Image, audio and text stay on this device. A clear, front-facing portrait works best. You align the face manually; no face identification service is used.

I am 18 or older. This is my own adult likeness, or the adult represented has explicitly permitted its use for generated animation. I have the right to use the image.I understand this is an artificial animation, and will not present it as an authentic recording of the person speaking.A local PNG, JPEG or WebP portrait, up to 10 MB and 24 megapixels

Agree above before choosing a portrait.

Horizontal positionVertical positionZoom

I want to download or reuse the local portrait animation model.

No portrait model is loaded. Generated speech is needed for portrait playback.

The still image helps you align the crop. Choose Prepare this portrait to run the neural model. Animation starts only after you choose Play generated speech with portrait. Timing is estimated from the text and can look unnatural.

Portrait software and model licences

Choose what to bring

A memory can stay beside you.

Choose a file, or continue with words.

Choose a local file

Optional. The file stays in this browser and is never passed to the local model. Up to 100 MB.

How would you like to use this? Talk beside a memoryKeep the recording as it is. Let the assistant ask about your memories and feelings. Personal questions show your own notes. Use my own likenessUse your own image beside the conversation. Notes stay in the app and are not used to create a personality. Name for this memory or session

Three notes you want to keep here

Use your own words. These notes stay in the app. They are shown as your own words for personal questions and are not passed to the model. At least one is needed.

First memory or approved fact Second memory or approved fact (optional) Third memory or approved fact (optional) I am 18 or older. I understand that replies are generated by a language model on this device. The assistant cannot know what this person would think or say, and may get things wrong.

Talk, at your own pace

The model has not been started.

Generated reply

Replies appear here, separate from your recording.

No need to rush.

Choose what to bring, then start when you feel ready. You decide when to return and when to stop.

Open saved conversation

For reflection, the assistant receives your current message and up to two recent exchanges. Your name and the separate notes above are kept for the app to show, without passing them to the model. Nothing is saved automatically. A saved conversation contains the personal text you entered and the generated replies. Your photo or recording is not included. Delete and start again clears this tab; it cannot delete files you downloaded.