txt2img means text-to-image: a diffusion model begins from noise and uses your prompt to guide repeated denoising toward a new image. On a phone, model choice, resolution, steps, guidance, and seed must fit the app and device.
The short answer
txt2img is the basic text-to-image workflow in Stable Diffusion. You provide a text prompt; the system encodes that prompt, starts from a noisy latent representation, and denoises it over a sequence of steps. A decoder turns the final latent into visible pixels. No source photograph is required.
The name describes the input and output, not where generation runs. A txt2img interface on a phone may run the model on the phone, on your computer, or on a cloud server.
The controls that change a txt2img result
The prompt describes desired content. The negative prompt can steer away from specified content when the pipeline supports guidance. The seed initializes the random process. Image width and height set the output shape. The scheduler determines how denoising steps are arranged, while the step count controls how many are performed.
Guidance controls how strongly the result is pushed toward the prompt in pipelines that use classifier-free guidance. Higher is not always better: Diffusers documentation notes that excessive guidance can reduce image quality or introduce artifacts. Start with the app’s model-specific preset instead of importing a desktop number without context.
How to run a useful first test
Choose one model, a moderate image size, and the balanced or recommended preset. Write a prompt with one subject, one action, a clear environment, a medium, light, and framing. Generate one image and save its seed and settings.
For the second image, keep the seed and every setting fixed. Change only one visible instruction—such as ‘close-up’ to ‘wide shot’—and compare the outputs. This tells you whether the model understood the change. Randomizing the seed at the same time turns the comparison into guesswork.
What a seed can and cannot reproduce
Within the same pipeline, a fixed seed and fixed settings can make iteration more reproducible. The CompVis reference sampler exposes a seed for this reason, and Diffusers supports deterministic generators.
Exact reproduction can still change across model files, app versions, schedulers, numerical precision, hardware paths, or implementations. Save the image itself alongside its settings. Treat the seed as workflow metadata, not as a permanent address for the pixels.
PhoneDiffusion availability and article scope
PhoneDiffusion provides text-to-image on supported iPhone and iPad devices running iOS 17 or later, using compatible curated model packs installed through the app. It also provides separate supported editing resources for image-to-image, inpainting, 4× upscaling, and background removal.
The exact settings depend on the installed pack and current release. PhoneDiffusion does not promise that a desktop recipe, checkpoint, scheduler name, or seed will reproduce identical pixels on iPhone.
Field notes
- txt2img describes a prompt-to-new-image workflow; it does not prove that generation is on-device.
- Prompt, seed, size, scheduler, steps, and guidance have distinct jobs.
- Use a fixed seed for controlled iteration, but save the output because cross-runtime reproduction is not guaranteed.
Try it on your phone
- Select one installed model and its recommended preset.
- Write a prompt with a specific subject, action, environment, medium, light, and framing.
- Generate once and record the prompt, seed, size, and preset.
- Keep everything fixed, change one visible phrase, and compare the pair.
Sources and verification
Primary and maintainer documentation checked for this guide. Product-specific details were reviewed against the current PhoneDiffusion scope.
- CompVis — Stable Diffusion reference implementationOriginal project documentation for latent text-to-image generation, prompts, seeds, dimensions, steps, and model limitations.
- Hugging Face Diffusers — text-to-image pipelineMaintainer documentation for prompt guidance, negative prompts, deterministic seeds, and text-to-image controls.