Troubleshooting¶

Find your symptom below. If it is not here, or the fix does not work, please open an issue with the run log attached — the copy button on the log above Run copies it — and your phone model.

A flow is dimmed and will not open¶

It needs a newer chip than your phone has; the card says which (for example "this flow needs v79 (Snapdragon 8 Elite or newer)"). Image edit and both video flows need an 8 Elite. Nothing can change this — the models are built for that generation of NPU.

A model says "this device cannot run it"¶

Same reason: the model family needs a newer chip. See the table on the home page.

A download is slow, stuck or fails¶

Check you are on Wi-Fi and have enough free storage (the row shows the size).
If Hugging Face is slow or blocked where you are, switch Settings → Downloads → Download source to hf-mirror.com.
An interrupted download resumes. A model left incomplete shows Repair, which fetches only the missing file.

Run does nothing, or says there is no model¶

Check the generator node has a Checkpoint chosen, and that the model is installed (Models). A flow opened with a model you have since deleted needs a new one picked.
The first Run after choosing a model loads it; watch the run log above Run.
The first Run of FLUX.2, Z-Image or Qwen Image also downloads the 23 MB engine they share.

The app closes (or restarts) during a render¶

Android closed it to free memory. In order:

Close other apps and try again.
Try a smaller size.
For SDXL and Anima, make sure low RAM mode is on in Settings → General → Memory.
For a FLUX.2 edit with a reference picture, use 512 × 512.
On Samsung phones, allow the app to run in the background (Settings → General).

A render stops when I switch apps¶

Allow background running: Settings → General → Keep rendering in the background. A render you leave running finishes; a model you are not using is let go a minute after you leave, which is normal (why). Swiping the app away from recent apps closes it, and stops a render.

RAM stays in use after I close the app¶

Update to 1.6.053 or newer. Before it, swiping the app away kept the model loaded until a timer that never fired, and only force stop freed it. Now a closed app lets the model go at once (when models are let go).

The first Run after coming back is slower¶

That is the model loading again after it was let go to give your phone its memory back. It takes a few seconds, once.

A node is red¶

The node failed; the reason is written underneath it. Everything after it is blocked. Common ones:

no photo chosen on an Image node that a generator needs;
an Inpaint node with nothing painted in its mask;
the backend stopping mid-render — usually memory, see above.

Image to image or an edit changes too much, or too little¶

Change Denoise: lower keeps more of the photo, higher changes more. Image to image opens at 0.65; an edit at 1.0. For inpainting, also check the mask covers what you want changed — and a little beyond it.

The picture ignores part of my prompt¶

SD 1.5 and SDXL read a limited number of tokens (about 75 for most models); the counter under the prompt box shows how much you have used. Put the important words first.
FLUX.2, Z-Image and Qwen Image read full sentences, and ignore the negative prompt at CFG 1. Say what you want instead of what you do not.
A Russian or Chinese prompt should be translated first (the 文A button) for SD 1.5 and SDXL.

Video flows say the models are missing¶

Install them under Models → Video (8.6 GB). They are separate from picture models.

Reporting a problem¶

Include: your phone model, the app version (shown next to the logo), what you did, and the run log (copy button on the log). A screenshot of the canvas helps too.