How does img2img work with a reference image?

Img2img combines the visual information in your picture with a written instruction to generate another image. The reference supplies composition, objects and perspective, while the prompt describes the change. Imgraft uses Subtle, Balanced and Bold instructions to guide preservation; these are prompt templates rather than numeric diffusion parameters, and small details may still change.

Is img2img free to try online?

Yes. Imgraft provides free use with no sign up or card; free use is limited and may be unavailable at busy times. Standard generation at 1K with a watermark is free; paid HD generation supports up to 4K without the watermark. All requests use moderation. A reference image does not guarantee an exact copy of its composition, so compare the generated version and keep your original alongside it.

Can AI edit only one part of an image without changing the rest?

A prompt describes which part to change; a mask selects an editing region; pixel preservation means leaving the original pixels outside that region untouched. These are different controls. Imgraft uses whole-image reference editing with no mask locking or masking brush, so other parts of the image may change even if your prompt asks to keep them. For an exact local correction, use an editor with region selection and verify how it preserves pixels outside the selection.

Img2img vs text to image: when should I use each?

Use img2img when you already have a composition, product arrangement, room or landscape to guide the result. Text to image starts from words and must invent that structure. A reference narrows what you need to describe, but the model still generates new pixels; use conventional editing when accurate labels, measurements or exact regions matter.

Can I use img2img without changing a face?

You can ask the model to preserve identity and use Subtle to request closer structure, but generated details can still change. Imgraft does not promise exact facial preservation or provide a face-swap function. For outfit edits, use the dedicated adult-consent workflow and discard any result that changes identity; do not use generated images as evidence.

Which img2img model should I choose on Imgraft?

Standard provides free 1K trials with a watermark; HD provides paid output up to 4K without it. Its sizes cost 1 credit at 1K, 2 credits at 2K, 3 credits at 4K. Compare the same source and prompt rather than assuming resolution alone improves accuracy.

How does img2img denoising strength work on Imgraft?

Choose Subtle for instructions that retain structure closely, Balanced for a recognizable layout, or Bold for a freer interpretation. Name the objects and camera angle to preserve. Imgraft implements strength through prompt wording; it does not expose or claim a diffusion denoising percentage. Return to the original if a refinement compounds unwanted changes.

What happens when an img2img request is rejected?

Required checks must approve the prompt, input images and final output before delivery. A blocked or confirmed failed generation returns your free try or credits; if we cannot confirm the result yet, your credits stay on hold until we can. Safety rules apply to paid and free models alike. Contact support@imgraft.com with the task ID if you think a decision was mistaken.

What Is Img2Img? How Image to Image AI Works

What is img2img (image2image)?

Img2img, also written image2image, generates a new image from a reference picture and an instruction. You might upload a greenhouse photograph and ask for a watercolor version, or upload a room photo and ask for different materials. The source gives the model visual information you would otherwise have to describe.

Imgraft at imgraft.com is an independent AI photo editor; About Imgraft identifies the product and its support contact.

Try the image to image AI workbench with a source you own; it works as a free img to img editor without sign up. The output is generated again rather than edited pixel by pixel. Objects can move, text may change and details can be invented. Keep the original and inspect the result when accuracy matters.

Img2img vs text to image

Text to image invents a scene from your description. Img2img starts with a scene you supply. If the camera angle, building shape or product arrangement already exists, a reference image can make the request easier to express. It does not guarantee an exact copy of those elements.

Try free img2img editing when you have a source. Use a text-only generator for a new composition. Use image to image AI when you want a different treatment of an existing picture. Use a conventional editor when you need to change one pixel region with predictable, exact results.

Which approach fits your task
Task	Use	Why
Change an existing picture	Img2img	Your picture supplies the layout
Start from nothing	Text to image	Words describe the whole scene
Fix one exact region	Conventional editor	Predictable, pixel-level control

Img2img denoising strength: what this tool actually controls

Some diffusion tools expose a numeric denoising parameter. Imgraft uses Subtle, Balanced and Bold prompt templates instead. Subtle asks the model to preserve structure closely. Balanced keeps the main layout recognizable. Bold asks for a freer interpretation. These are written instructions, not a hidden denoise percentage.

The underlying idea is described in SDEdit, a research method that adds noise to an input image and then denoises it, so the output stays related to the source.

subtle room restyle from the same living-room photograph — SUBTLE

balanced room restyle from the same living-room photograph — BALANCED

bold room restyle from the same living-room photograph — BOLD

If furniture positions matter, start with Subtle and name the fixed elements. Bold may be useful for exploring a different visual world, but it is less suitable for documenting a real room.

What a reference image does

The reference supplies shapes, colors, perspective and relationships between objects. A clear source usually gives you more useful constraints than a low-resolution picture with an ambiguous subject. Upload one JPG, PNG or WebP up to 10 MB. Imgraft resizes its longest edge to 2048 pixels before generation.

A standard edit uses one source; the AI clothes changer also accepts a garment reference for authorized adult photos. For line-based treatments, try image to sketch. For a scratched or faded print, start with AI photo restoration. To continue an edit, choose Refine this: the current result becomes the next reference. Each step is a separate generation, and each can introduce small differences.

Match the model to the output you need

Standard vs HD generation on Imgraft
Mode	Best for	Output size	Generation credits	Watermark
Standard	Everyday edits	1K	Free try	Yes
HD	Detailed restyling	1K	1	No
HD	Detailed restyling	2K	2	No
HD	Detailed restyling	4K	3	No

Try an edit for free with Standard, or choose HD for fuller detail and larger downloads. Start at 1K to compare the treatment with your source. Once you have a direction you like, select an output size for your presentation or layout. The table shows the credit cost beside each tier before you generate.

Imgraft is an independent product built on third-party AI models. See the AI usage policy for provider details. All results are AI-generated.

Compare model options and credit costs before choosing your output size.

How to Edit an Existing Image With AI Using Text Prompts

Upload a clear image you own or have permission to edit.
Write a custom prompt naming the change and the details to keep, or choose a style and add optional details. View full prompt shows the style instruction; Edit as custom makes the combined text editable.
Confirm permission and generate. The model creates a new image from your reference and instructions.
Compare the result with the original using the slider. Check shapes, faces and text before downloading or choosing Refine this.

Prompt example: "Change the room walls to warm white plaster; keep the window positions, sofa and camera angle." Avoid conflicting instructions, such as replacing a room while keeping every object untouched.

Preservation instructions are requests, not locks. Objects, lettering and other details may change. Return to the original if a refinement carries an unwanted change forward; use a conventional editor for pixel-exact corrections.

For product shots, the AI background changer starts with prompts asking to keep the product and replace the scene. If you need words for a reference first, describe an image as a prompt and check the generated text before editing.

Six prompt examples to start with

Restyle

Turn this photo into a loose watercolor painting on cold-press paper, soft washes, visible paper texture. Keep the exact layout.

Replace background

Place this object on a clean studio backdrop with soft side lighting and a gentle contact shadow.

Change season

Make it a snowy winter evening, warm light in the windows. Keep the same scene.

Restore old photo

Remove scratches, dust and fading. Add natural color while keeping the original scene.

Product studio shot

Place this product on a marble pedestal in a minimal studio, warm rim light, soft contact shadow.

Sketch to render

Turn this architectural sketch into a realistic render at golden hour, keeping the same design and angle.

Common failures and fixes

The result changes too much.

Return to the original, choose Subtle and name the objects that should keep their position. Refining an already changed result can carry those changes forward.

The style is too weak.

Name materials and marks you can see: cold-press paper, loose watercolor washes or graphite hatching. Try Balanced before adding several competing styles.

Small details look wrong.

Larger output is not a repair guarantee. Use a clearer source, narrow the request and check the result with the comparison slider. Product labels and architectural dimensions need manual review.

A request is rejected.

Safety checks apply to every plan. Do not try to bypass them. Face manipulation and adult content are prohibited; use the support address in the Acceptable Use Policy if you think a review was mistaken.

What Is Img2Img? How Image to Image AI Actually Works