Crochet Filter Not Working? Fixes for Common Problems
The crochet filter usually works on the first or second try, but when it doesn't, the cause is almost always one of a handful of problems. Find yours below. Each fix works in any of the free tools in our comparison: ChatGPT, Gemini, Microsoft Copilot or Vheer.
A general rule first: change one thing per retry. Try a better photo, or a prompt tweak, but not both at once, so you know what helped. And in chat-based tools like ChatGPT and Gemini, a short correction in the same chat often works better than a brand-new prompt.
The doll doesn't look like me
This is the most common complaint. Fixes, in order of how much they help:
- Use a better photo. Face large in the frame, facing the camera, even front light, no sunglasses, no hand on the face.
- Crop before uploading. If you're small in a wide shot, crop to head and shoulders or waist up.
- Name your key features. Add one line: "keep my curly dark hair, thick eyebrows, round glasses and the mole on my cheek."
- Ask for less cartoon. Some prompts push big doll eyes so hard that likeness disappears. Try "keep my real face shape; smaller doll eyes."
- Try another tool. Different models keep faces differently. The same photo can work in one and not the other.
The face stayed a real face with a knitted texture
You wanted a doll and got a person in a yarn costume. Reply with: "The face must be a crocheted doll face made of yarn stitches, not a photo face." A photo where your face isn't tiny also helps.
Only part of the scene turned into crochet
If the person is yarn but the room is half real, say: "Every wall, floor, piece of furniture and object must also be crocheted." Simpler backgrounds convert more evenly. If you actually prefer a real background, ask for that explicitly instead: "Keep the background as a real photograph."
It looks like a plush toy or felt, not crochet
Add the words that describe crochet specifically: "visible single-crochet stitches", "yarn texture", "amigurumi". "Knitted" alone sometimes gives a sweater texture instead of a crochet doll.
Text, logos or writing came out garbled
AI models often mangle text. Ask for "any logo or writing becomes plain knitted fabric", or choose a photo without big logos.
Two people blended together, or someone vanished
See the couples and family guide. The short version: faces apart in the photo, and say exactly how many dolls you want.
The tool refused my photo
AI tools have content rules. Photos of children, public figures, or anything that looks like it could be someone who didn't consent may be declined. Use a photo of yourself or adults who agreed. If a normal selfie is refused, try again later or with a different photo; refusals are sometimes inconsistent.
The picture came out cropped or the wrong shape
Add the format you want: "vertical 9:16" for TikTok and Reels, "square" for a profile picture, or "keep the original framing".
It's blurry or low resolution
Upload the original photo file, not a screenshot or a compressed copy from a messaging app. Some free plans output at a lower resolution than paid ones; check the tool's own page.
"You've reached your limit"
Free plans cap how many images you can make. ChatGPT tells you when your limit resets; Google says Gemini's limits refresh after some hours. Meanwhile, the limits are separate per tool, so you can continue in another one with the same prompt.
Nothing happens, or it says something went wrong
Image generation can take a minute or more, especially at busy times. Wait, then refresh. If it keeps failing, start a new chat, upload the photo again and paste the prompt.
Still stuck?
Start again with the shortest working prompt: "Turn this photo into a handmade crochet amigurumi doll. Keep me recognizable." Once that works, add details back one at a time. Tool-specific steps: ChatGPT, Gemini.