SimpleSDImage FAQ
Common questions about SimpleSDImage, with answers.
Images only come out blurry
This happens when a normal (non-LCM) model is added with LCM model turned on. LCM draws in only 5 steps, which is not enough for a normal model to finish the image.
In Config, tap Change model and add the same model again with LCM model turned off. LCM model can only be set when you add a model.
Drawing is slow and images come out harsh or broken
This happens when an LCM model is added with LCM model turned off. The app then draws it the way it draws normal models (20 steps, following the prompt strongly), which is 4 to 8 times slower and makes an LCM model's images too intense or broken.
LCM model is off by default. For models with "LCM" in the name, turn it on before you add the model.
To check, look at the model name in Config: an LCM model shows "(LCM)" after its name. If it doesn't, tap Change model and add the model again with LCM model turned on.
What's good about LCM models?
Speed. LCM models finish an image in 5 steps, so they draw 4 to 8 times faster than normal models, which need 20 steps. On a phone, where drawing is slow, this makes a big difference. The built-in model (DreamShaper 8 LCM) is an LCM model.
With an LCM model, leaving the negative prompt empty makes drawing about twice as fast again. Normal models always take what you don't want into account, so they can't use this shortcut.
On the other hand, normal models often draw finer detail, and negative prompts work better with them. If you care more about the result than the time, a normal model is a good choice.
The LCM model checkbox must match the model you add. Turning it on for a normal model doesn't make it faster; it only makes the images blurry. To find LCM models, look for "LCM" in the model name.
What prompts should I use?
List what you want as short words, separated by commas. Put the most important things first: what to draw, how it looks, then the style. Here are some prompts you can use as is.
Landscape: a quiet lake in the mountains at sunrise, mist, soft light, highly detailed, landscape photo
Person: portrait of a young woman, smiling, long brown hair, natural light, blurred background, photo
Illustration: a cute cat wearing a wizard hat, anime style, colorful, simple background
Food: a bowl of ramen on a wooden table, steam, close-up, food photography
Room or building: a cozy cafe interior, warm lighting, plants, wooden furniture, digital painting
Negative prompt (what you don't want), for example: blurry, low quality, bad hands, extra fingers, text, watermark. With LCM models, leaving it empty draws about twice as fast.
The same prompt gives a different image each time, so draw a few times until you like one. Packing too many things into one image makes them all more likely to come out broken.
Which size should I choose?
You can choose 512×512 (square), 512×768 (portrait) or 768×512 (landscape). Pick the one that fits the shape of what you want to draw.
512×512: the fastest. Good for a single object, icons, or trying out a prompt.
512×768 (portrait): good for full-body people and phone wallpapers.
768×512 (landscape): good for scenery, streets, room interiors and other wide scenes.
Portrait and landscape have 1.5 times as many pixels, so they take longer. To wait less, settle on a prompt at 512×512 first, then draw it again in portrait or landscape.
Stable Diffusion 1.5 was trained on 512-pixel images, so larger sizes often draw people or objects twice. That's why the app offers only these three sizes.
I don't like a face in my image
Open the image in the Images tab and tap Fix faces. The app finds small faces, which often come out broken. Tap the faces to fix, then tap Fix selected faces.
If no face is found, or you want to change a large face, use Edit. Paint over the face, enter the face you want (for example, smiling face), and tap Redraw painted area.
Both save the result as a new image and keep the original, so you can try as many times as you like.
Change faces only in images drawn by the app, or in photos where the people shown have agreed.
How do I remove trash or unwanted things from a photo?
In the Images tab, tap Upload to add the photo, open it, and tap Edit.
Paint over the thing to remove, a little beyond its edges. As the prompt, enter what should be there instead (for example, grass, blue sky, wooden floor), then tap Redraw painted area. Describe only what should be there, not the thing you want to remove.
If it doesn't go away, raise Strength so the result follows the original less. The result is saved as a new image and the original photo is kept.
Uploaded photos are resized so the long side is 768 pixels, and the edges may be trimmed slightly.
Use photos you took yourself or have permission to use. For photos of people, get their agreement first and keep your edits true to what really happened.
How do I change the model?
Open Config and tap Change model. There are two ways to add a model.
From a file on your device: tap Choose file and pick a single model file (.safetensors, .ckpt or .gguf).
From the web: paste the file's download link or the Hugging Face model page link, then tap Download. Models split into folders (unet/, vae/ …) can only be added this way.
Either way, turn on LCM model if the model is an LCM model, and leave it off otherwise, before you start. After downloading, the app converts the model for your device automatically.
The app uses one model at a time, so a new model replaces the current one. Converting needs up to about 6GB of free space. Your images are kept.
To go back to the original model, tap Use default model on the same screen.
How do I find other models on Hugging Face?
The app works with Stable Diffusion 1.5 models. On Hugging Face (huggingface.co), search for "stable diffusion 1.5" or "sd 1.5". SDXL, SD 2.x, SD3 and FLUX models don't work.
If the model page's Files and versions tab shows folders such as unet, vae and text_encoder, paste the page link (https://huggingface.co/owner/model-name) in Config > Change model. If the model is a single file such as .safetensors, paste that file's download link (…/resolve/main/…).
Turn on LCM model only for models with "LCM" in the name. Leave it off for all others.
Models that need sign-in or agreeing to terms, and pages that only offer a LoRA, can't be used. Check each model's license before use.
Whatever model you use, keep what you draw suitable for all ages and within the law and the model's license.