Your phone can be the place you create without doing all the processing. OGAM (Off Grid AI Mobile) can send an image prompt to OGAD (Off Grid AI Desktop) on your own computer, then show the result on your phone. With a downloaded image model and a working local network, generation does not need internet.
What would you like to do with Off Grid AI?
Have a feature or use case you would like us to support? Tell us what you want to do and which device you use.
Write to support@offgridmobileai.co, join our Slack community, or talk to us on Reddit.
You can sketch an idea from the sofa, change a prompt while looking at a real object, or try several visual directions without returning to the computer for every request. The computer supplies the image model and its processing resources. The phone supplies the prompt and displays the result.
Initial app and model downloads need internet. Keep both devices connected to your own local network for the steps below.
Where does the image model run?
The image model runs on the computer you choose in OGAM. Your phone sends the prompt across the network and receives the generated image. You do not need to store that same image model on the phone.
This is different from moving a downloaded model between devices. You leave the model on the computer and use it remotely. It is also different from syncing an image that you already made: here, the request starts on your phone.
Use OGAM 0.0.111 with a compatible desktop release such as OGAD 0.0.51. The remote image path is part of the core model-server workflow; it does not require Pro device sync.
What do you need before you start?
Install OGAM on Android or iPhone. On your computer, install OGAD and download a supported local image model. Generate one image on the computer first so you know the model is ready.
You also need:
- A home network that lets the two devices reach each other.
- Enough memory and storage on the computer for the chosen image model.
- A computer that stays awake with OGAD running.
- Space on the phone to save the returned images.
A model that works on the computer is the important requirement. A remote connection cannot make an unsupported desktop model work. Use a downloaded local image model on the host if you want the whole generation path to stay on your hardware.
How do you connect the phone to the computer?
Add the computer under Settings → Remote Servers in OGAM. Test the connection and choose an image model the computer provides. OGAD's local gateway uses port 7878.
- Open Gateway in OGAD and keep the app running.
- Find the computer's local IP address in its network settings.
- On the phone, open Settings → Remote Servers and add a server.
- Enter a name such as Home images.
- Enter the address in the form
http://YOUR-COMPUTER-IP:7878. - Tap Test connection.
- Choose a downloaded model in Image model, then tap Add server.
For example, http://192.168.1.50:7878 would be correct only if 192.168.1.50 were your computer's address. Do not enter the phone's address or a public website.
The OGAD gateway in this setup does not require an API key. Keep it on a trusted network and restrict access through the computer's firewall. Do not expose port 7878 through public router forwarding.
If the phone requests local network access, allow it so it can reach your computer. Guest Wi-Fi can block device-to-device connections even when both devices appear to be online.
How do you create the first image from the phone?
Open OGAM's model selector, find the image models, and select the one listed under your computer's name. The Remote label distinguishes it from a model on the phone. Wait for selection to finish, then send one simple image request.
Try a prompt with a clear subject and style:
Generate an image of a small green reading lamp on a wooden desk, warm evening light, simple background, square composition.
For an explicit image request, use the chat's Image Gen control to choose image generation. This helps you get an image rather than a text answer about the subject.
During the request, OGAM can show generation progress received from OGAD. When it finishes, the image is saved on the phone and can appear with the conversation's generated-image result. Keep the phone connected until the image arrives.
Start with one image. Check whether the subject, lighting, and composition match your idea. For the next prompt, change one part:
Generate the same kind of desk scene with cool morning light and a white ceramic lamp.
This second prompt describes a new generation. It does not promise that every object or pixel will match the previous image. Reference-image editing is a separate workflow.
Can this work when the internet is down?
Yes, when both apps and the model are ready and the phone can still reach the computer over the local network. A router can provide local Wi-Fi even when its internet connection is unavailable.
Keep Wi-Fi connected on both devices. Turning off all network connections would stop the phone from sending the prompt and receiving the image.
For a useful check, complete one request with your normal connection. Then remove internet access while leaving the local network running and try another prompt. Make sure the chosen desktop model is local and fully downloaded before this check.
Access from a different location is another setup. A private remote connection such as Tailscale normally needs internet at both ends. It should not be described as this internet-free home-network workflow.
Why did the image request fail?
Check whether the computer can still generate an image locally. If that fails, fix the host model first. If local generation works, check the phone's server selection and network connection.
| What you see | What to do |
|---|---|
| No remote image model | Finish downloading the model on the computer and test the server connection again. |
| A text answer instead of an image | Select the remote image model and use the explicit Image Gen mode. |
| Server unavailable | Wake the computer, open OGAD, and check its current IP address. |
| A generation or memory error | Use a model and image size the computer can handle. Close other demanding work on the host. |
| Generation starts but the result never arrives | Keep both devices connected and OGAM open while the request completes. |
The host still needs enough memory for image generation. Sending the request from a phone does not remove that limit, and a larger computer does not guarantee a fixed generation time.
Where do the prompt and image go?
The prompt leaves the phone and goes to the selected computer. The computer runs the downloaded model, then sends the result back to OGAM. With your own local host, there is no required cloud image-generation service in this route.
If you choose a different remote server, its operator receives the request. If the home computer itself forwards work to a cloud model, that also changes where processing happens. Check the active model before sending a private idea.
Make one image with hardware you already own
Install OGAM on your phone and OGAD on your computer. Prepare one local image model, connect the phone, and try the reading-lamp prompt. Once it works, you can explore your next visual idea from the phone while your computer does the generation.
Top comments (0)