Local iPhone Stable Diffusion 1.5 Benchmark - ridiculously fast generations!

PhoneDiffusion — Tue, 02 Jun 2026 22:46:26 +0000

I have been testing how far Stable Diffusion 1.5 can be pushed running locally on iPhone 17 and how good the results are.

For this benchmark I used PhoneDiffusion with three SD 1.5 model packs:

CyberRealistic
DreamShaper 8 LCM
Realistic Vision V5.1 Hyper

They have different assumptions baked into their recommended settings, schedulers and step counts. That makes them a good small test set for local image generation, because they represent three different modes: quality pass, fast iteration and very fast draft generation.

The Benchmark

The test was intentionally simple.

I used three prompts:

a cyberpunk penthouse
an anime forest elder
a futuristic cybertruck

Each model generated three images per prompt. That produced 27 images total. For the final comparison, I selected the strongest visual result from each model/prompt pair, giving 9 selected outputs.

All images were generated at 512 x 512. The timings came from warm PhoneDiffusion generation metadata, with the model packs already installed and prepared. Compute used CPU + Neural Engine.

This is not a cold-start benchmark. It is closer to what a user sees after the app is open, the model is ready, and they are iterating on prompts.

Why the Step Counts Are Different
A common way to benchmark models is to force every model through the same settings.

That would make this comparison less useful.

CyberRealistic, DreamShaper LCM and Realistic Vision Hyper are meant to run differently. If every model were tested at the same step count and CFG, the result would mostly measure bad configuration choices.

Model Benchmark for iPhone 17

CyberRealistic behaves like the quality-oriented model in this group.

At 30 steps / CFG 7, it averaged 13.7s across the selected outputs. That is not instant, but the extra time shows up in the images. The cyberpunk room has more structure and surface detail. The cybertruck result looks more intentionally designed.

DreamShaper 8 LCM is the most balanced result.

At 10 steps / CFG 2, it averaged 4.7s on selected outputs. That is fast enough to keep trying prompts without losing momentum, while still producing images that are visually coherent. This is the one I would use for most exploration work.

Realistic Vision V5.1 Hyper is the speed configuration.

At 6 steps / CFG 1.5, it averaged 3.2s for selected outputs. That is a different interaction model. You can test prompt wording, composition, and subject placement quickly, then switch to a slower model when the direction is right.

If you want to try it out yourself you can learn how to run Stable Diffusion on your iPhone here:

Download Phonediffusion here: https://apps.apple.com/us/app/phonediffusion/id6762061991

PhoneDiffusion: Local AI Image Generation, Built for iPhone

PhoneDiffusion — Thu, 28 May 2026 21:29:09 +0000

AI image generation on mobile still feels harder than it should. Many apps rely on cloud rendering, credit systems, queues, accounts, and limited control. You write a prompt, spend a credit, wait for the result, and hope it matches what you had in mind.

PhoneDiffusion takes a different approach.

Generate AI images locally on your iPhone with PhoneDiffusion: https://apps.apple.com/us/app/phonediffusion/id6762061991

It is an iPhone app for local AI image generation, built around curated Stable Diffusion model packs converted for Apple Core ML. After downloading a compatible model pack, users can generate images directly on their device without sending prompts or images to a cloud service.

The goal is simple: make private, on-device image generation feel practical on iPhone.

PhoneDiffusion does not require an account. Prompts, negative prompts, generation history, images, and model files stay on the device unless the user chooses to export, share, or submit feedback. Instead of asking people to manage random model files manually, the app uses curated packs that are prepared for iOS.

The app currently focuses on two main areas: generation and gallery. In Generate, users choose a model, select a style, write a prompt, adjust settings if needed, and create an image. In Gallery, they can revisit previous generations, compare results, save, share, or delete them.

Under the hood, PhoneDiffusion uses Apple’s Core ML Stable Diffusion stack. The launch model set includes fast SD 1.5-class options and heavier SDXL Lightning-class models, including DreamShaper, Realistic Vision, CyberRealistic, RealVisXL, and other mobile-oriented experiments.

This is not meant to replace a full desktop Stable Diffusion workstation. Power users already have excellent tools for advanced workflows such as LoRAs, ControlNet, inpainting, outpainting, custom model imports, and complex graph-based workflows. PhoneDiffusion is aimed at a more focused use case: private, curated, phone-native image generation that is approachable enough for regular creators, while still useful for people who understand Stable Diffusion.

The basic flow is straightforward. Install the app, download a compatible model pack, let the app prepare the local pipeline, enter a prompt, generate, and save the result locally.

A major part of the work was performance testing. Local image generation on iPhone is not just a matter of packaging a model and pressing run. Model downloads can be large, iOS memory pressure matters, devices heat up, and Core ML execution behavior can vary depending on device state.

In testing, SD 1.5-class models proved to be the best path for fast and reliable first results. SDXL can deliver better image quality, but it comes with heavier cold-start, memory, and thermal tradeoffs. For many users, especially without the newest Pro-level iPhone hardware, the faster SD 1.5 models are likely to feel better in day-to-day use.

The team also benchmarked image quality manually, looking at visual quality, prompt adherence, and artifacts. Automated scores can help, but for a creative tool, human judgment still matters. If a model misses the subject, layout, or product details, the user notices immediately.

PhoneDiffusion is built for people who want private visual brainstorming on iPhone without a desktop setup or cloud workflow. Useful early workflows include avatars, tattoo ideas, thumbnails, poster concepts, product-shot drafts, interior moodboards, wallpapers, character ideas, and quick style exploration.

The broader idea behind PhoneDiffusion is that mobile AI should not simply copy desktop AI tools at a smaller size. Phones have different constraints, different workflows, and different expectations. A good local AI image generator for iPhone needs to respect those limits while still giving creators meaningful control.

You can try PhoneDiffusion here: https://apps.apple.com/us/app/phonediffusion/id6762061991

DEV Community: PhoneDiffusion

Local iPhone Stable Diffusion 1.5 Benchmark - ridiculously fast generations!

The Benchmark

Model Benchmark for iPhone 17

PhoneDiffusion: Local AI Image Generation, Built for iPhone