Create Image

Summary#

Create Image takes a reference picture, describes what it sees, and turns that description into a new original image. You upload a reference, Claude breaks it down into composition, subject, pose, clothing, lighting, colour, background, camera angle and style, and writes a generation prompt from that analysis. You can edit the prompt, choose how closely to follow the reference, and generate.

Purpose#

Recreating a visual concept usually means writing a prompt from scratch and guessing at the words a generator responds to. This tool reads an image you already like and writes that prompt for you, so you start from an accurate description instead of a blank box.

Overview#

The tool works in two steps, and they are charged separately.

Analyse sends your reference to Claude, which returns nine short observations and one detailed prompt of roughly 60 to 120 words. Nothing is generated at this stage.

Generate sends the prompt — and only the prompt — to the image generator. Your reference image is never passed to it. The new image is built from a written description, which is why the result is a new picture in a similar spirit rather than a copy of your upload.

Generated images are saved to your media library, so they can be reused elsewhere without generating again.

Benefits#

• Starts you from a detailed, accurate prompt instead of a blank field.
• Shows you what it saw, so you can correct the description before spending credits on generation.
• The prompt is editable, so the tool is a starting point rather than a black box.
• Two similarity modes cover both "something like this" and "this, but change the jacket".
• Results land in the media library and can be reused.

Use Cases#

• Recreating the look of a photograph you have the rights to reference, as an original image.
• Exploring variations on a concept: same composition, different subject, clothing, lighting or season.
• Turning a mood board into usable art direction, because the analysis is written in the language image generators respond to.
• Producing a matching image when you have one that works and need a second.

Requirements#

• A paid plan. The free technical SEO tools stay free, but both steps of this tool cost credits and are refused on the Free plan.
• A signed-in account.
• A reference image in PNG, JPEG, WebP or GIF, under about 4MB.

Permissions#

Any signed-in user on a paid plan can use it. There is no separate role or admin permission. Generated images are saved to your own media library and are not shared with other accounts.

Cost#

Analysing and generating are billed apart on purpose. If you read the prompt and decide against it, you have spent 1 credit rather than 5. Generation is charged per image actually returned, so a request for four that comes back with three is charged for three.

Generating four images in one press costs 20 credits: 4 for the analysis run plus 16, or 16 alone if you reuse a prompt you already have.

Navigation Path#

Dashboard > Image Tools > Create Image

Inputs#

Step-by-Step Guide#

1. Open Image Tools, then Create Image.
2. Drop a reference image onto the panel, or press Choose an image.
3. Press Analyse reference. The breakdown appears on the right and the prompt fills in below.
4. Read the prompt. Edit it if the description missed something or emphasised the wrong thing.
5. Choose a similarity mode. Closer to the reference keeps composition, lighting and style. Refine the subject keeps the concept and changes what you describe.
6. In Refine mode, type what should change, or tap the chips for clothing, hairstyle, pose, expression, background, lighting or colours. Chips add to what is already there, so several can be combined.
7. Pick a shape and how many images you want. The credit cost updates as you change it.
8. Press Generate image.

Reading the Results#

The analysis panel lists nine observations. They are what the prompt was built from, so if one is wrong, editing the prompt is usually quicker than uploading a different reference.

Generated images appear below with a Download link and an Open link. They are already in your media library.

Examples#

A portrait reference, Closer mode. The analysis reports a three-quarter turn, soft window light from the left, a muted grey palette and a blurred street background. Generating produces a new portrait with the same framing and light, and a different person.

The same reference, Refine mode, "a red raincoat, evening light". The composition and camera stay; the coat and the light change.

Screenshots#

Not included. The tool is three panels on one screen and the step-by-step guide above follows them in order.

Tips#

• Analyse once, then generate several times with different edits to the prompt. Only the generation is charged again.
• The prompt is the whole input. If the result is not what you wanted, change the words rather than the reference.
• Refine mode understands plain descriptions. "Looking away, softer light" works as well as anything more formal.
• Use 1 image while you are still adjusting the prompt, and raise the count once it reads correctly.

Best Practices#

• Reference images you have the rights to use.
• Check the analysis before generating. It is the cheapest point to catch a misreading.
• Keep the edited prompt if you like the result, so you can regenerate without paying to analyse again.

Common Mistakes#

• Expecting a copy. The tool makes a new image from a description, not a reproduction of the upload.
• Generating four images before the prompt is right, which costs 16 credits to learn something 4 would have told you.
• Uploading a very large file. Anything over about 4MB is refused; resize it first with Image Resize.

Limitations#

• The reference is never sent to the image generator, so the result will not match it pixel for pixel. That is deliberate.
• The analysis describes people generically, by apparent age range, clothing and posture. It does not identify anyone and does not describe a face in enough detail to reproduce it, so the tool cannot be used to recreate a specific person's likeness.
• Recognisable artworks, posters and stills are described by their generic visual qualities only, never by name or author.
• Maximum four images per press.
• Image generation depends on an external provider. When it is unavailable the tool says so rather than returning a lower-quality substitute.

Troubleshooting#

"This tool needs a paid plan." Both steps cost credits, so the Free plan is refused. The ten free technical SEO tools are unaffected.

"That image is too large." Use a file under about 4MB. Image Resize will bring most photographs under that in one step.

"The image could not be analysed just now." The analysis provider did not answer, or answered in a shape the tool could not read. Nothing is charged for a failed analysis. Try again.

"Cover image generation is temporarily unavailable." The image provider is down. Your prompt is still in the box, so you can generate when it returns.

Fewer images came back than I asked for. The generator retries failures internally and returns what it managed. You are charged only for what arrived.

FAQs#

Does this copy the image I upload? No. Your reference is analysed into a written description, and only that description is sent to the generator. The output is a new image.

Can I recreate a photo of a specific person? No. The analysis deliberately describes people generically and does not capture identifying facial detail, so the tool cannot reproduce a particular person's likeness.

Can I use my own prompt without uploading anything? The prompt box is editable, but the tool is built around a reference and the Analyse step needs an image. Blog Studio's cover generator takes a prompt on its own.

Where do the images go? Your media library, reachable from Blog Studio, and downloadable from the result panel.

Does editing the prompt cost anything? No. Only Analyse and Generate are charged.