How to Use Imagine Image 2.0

Published August 8, 2026 · Updated August 8, 2026 · By Imagine2 Editorial

Imagine2 provides generation, editing, and reference workflows in a focused browser workspace. You can begin with a written direction, refine an existing image, or combine multiple references while keeping the prompt, settings, and credit estimate visible.

This guide explains how to use those workflows deliberately, from preparing a prompt or reference through reviewing, saving, remixing, and deleting a result.

If you are asking how to use Imagine Image 2.0 for practical creative work, the same discipline applies: choose the right starting mode, make constraints visible, and review the result against the brief rather than treating generation as a one-click finish.

This guide covers the complete V1 workflow. It explains what to write, how to prepare references, how credits are calculated, what happens during Google sign-in, how results are retained, and how to use History for remixing and editing. It also explains the limits you should know before putting an output into production.

Before you start

Imagine2 uses the xAI model grok-imagine-image-quality. The service is independent and is not affiliated with or endorsed by xAI. The model is generative: it interprets your direction rather than executing a deterministic graphics command. Good results come from a clear brief, appropriate references, and critical review.

You can begin on the homepage or any public tool page without signing in. Type a prompt, choose settings, and select local reference files. Imagine2 does not upload those files merely because you selected them. When you choose Generate while signed out, the browser stores a draft in IndexedDB, sends you to Google sign-in, restores the draft on return, and asks you to confirm again. It does not automatically spend credits after authentication.

That confirmation step is deliberate. It gives you a final opportunity to review the prompt, file selection, resolution, output count, and estimated cost.

Choose the right mode

Create: text to image

Use Create when the desired image can be described without visual source material. The mode accepts no references. It is appropriate for concept exploration, product still-life directions, editorial scenes, environments, material studies, and campaign ideas.

Create gives the model the most freedom. That freedom can be useful early, but it makes prompt structure more important. If you need a specific person's identity, exact product shape, existing room, or established visual asset, use Edit or Multi instead.

Edit: one source image

Use Edit when one image establishes the subject or composition and language describes the transformation. Edit requires exactly one JPEG, PNG, or WebP reference up to 10 MB.

Common tasks include changing time of day, replacing a background, adjusting clothing, transforming visual style, modifying product material, cleaning a scene, or creating a campaign variation. State what must stay unchanged as well as what should change.

Multi: two or three references

Use Multi when no single image contains all the visual evidence needed. The mode requires two or three references. Examples include combining a person with a wardrobe reference, placing a product into a location, applying one material to another object's form, or assembling subjects into a shared scene.

Assign each reference a role in the prompt. Do not assume the model knows which image supplies identity and which supplies style. “Image one is the subject, image two is the jacket, and image three is the location” is much clearer than “combine these.”

Build a useful creative direction

A prompt is not only a list of attractive adjectives. It is a hierarchy of decisions. A practical structure is:

1. Subject: What must appear?
2. Action or state: What is happening?
3. Composition: Where is the subject in the frame and from what viewpoint?
4. Light: What direction, quality, color, and time does the light imply?
5. Material and detail: Which surfaces, textures, construction, or camera cues matter?
6. Mood and palette: What should the image feel like without replacing concrete direction?
7. Invariants: What must remain unchanged in an edit?
8. Exclusions: What must not appear?

Consider a weak prompt: “Make a luxury perfume ad.” It leaves product shape, camera, environment, palette, surface, lighting, and layout open.

A stronger direction is: “Unbranded rectangular perfume bottle in smoked glass, front three-quarter product view on polished black stone, one hard cyan rim light from camera left, restrained violet reflection, precise studio photography, generous negative space above, no flowers, hands, labels, or typography.”

The stronger prompt does not guarantee perfection. It makes success measurable. You can inspect whether the bottle is rectangular, the material is smoked glass, the view is correct, the light is directional, the composition leaves space, and unwanted elements are absent.

Write better edit instructions

An edit prompt should separate preservation from transformation. Use phrases such as “keep,” “preserve,” “do not change,” and “replace only.” This reduces ambiguity when the source contains valuable identity or geometry.

For example:

“Keep the subject's face, expression, pose, and camera crop unchanged. Replace only the studio background with a rainy Tokyo street at blue hour. Match the new environment light on the subject, preserve realistic skin texture, and add no text.”

If the requested transformation changes the physical light, asking the subject to remain pixel-identical may conflict with realism. Decide which constraint has priority. “Preserve identity and pose while adapting light naturally” is a more coherent instruction than demanding an unchanged subject inside a radically different environment.

Prepare reference images

Use the cleanest source you are authorized to process. Higher resolution can help, but relevance and clarity matter more than a huge file. Avoid screenshots with interface chrome, heavy compression, watermarks, unrelated borders, and multiple ambiguous subjects.

For identity preservation, choose a source with a clear face, appropriate angle, and enough detail. For product form, show the silhouette and distinctive construction. For material transfer, use a reference where texture, scale, and reflectivity are visible. For locations, choose an image whose perspective can plausibly accommodate the subject.

Imagine2 accepts JPEG, PNG, and WebP up to 10 MB each. Selecting a file creates only a local preview until a signed-in user confirms generation. Uploaded assets are private and are accessed through authenticated ownership checks.

You are responsible for rights and consent. Do not upload a person's image without a lawful basis and any required permission. Do not upload confidential, highly sensitive, or illegally obtained material.

Choose an aspect ratio

The frame influences composition before the model places the subject. Choose it based on the final use rather than cropping everything later.

• 1:1 works for product tiles, profile concepts, catalog imagery, and flexible feeds.
• 3:2 is a familiar photographic landscape frame for editorial and environmental scenes.
• 2:3 suits full-length portraits, posters, and vertical product compositions.
• 4:3 provides a slightly calmer landscape frame for interiors, documentary work, and presentations.
• 3:4 suits portraits and editorial covers without the extreme height of mobile video.
• 16:9 suits wide campaign images, headers, scenes, and presentation backgrounds.
• 9:16 suits vertical mobile placements and tall subject compositions.

If typography will be added later in a layout tool, ask for negative space in a specific region. Do not rely on the model to render final brand copy correctly.

Choose 1K or 2K

Use 1K while exploring. It costs fewer credits, arrives faster, and is sufficient for reviewing composition, palette, and broad detail. Use 2K after the creative direction stabilizes, when the final placement is larger, or when you need room for a crop.

Resolution does not repair a weak prompt. A 2K image can express the wrong idea in more pixels. A productive workflow is to settle subject, frame, and light at 1K, then request a 2K variation using the refined direction.

In V1, each 1K output costs 5 credits and each 2K output costs 7. Each reference costs 1 credit. The workspace calculates the estimate before you submit.

Choose one, two, or four outputs

One output is best when the instruction and references are already constrained. Two outputs provide a modest comparison. Four outputs are useful when exploring composition or styling within a stable brief.

Avoid using four outputs to compensate for an unclear prompt. If the contact sheet varies in all the wrong ways, revise the direction before spending on another large batch. The goal is not the highest image count. The goal is a small set you can evaluate against the brief.

Understand the credit estimate

Imagine2 calculates:

output cost × output count + reference count

One 1K Create request costs 5 credits. Two 2K outputs from one reference cost 15 credits. Four 1K outputs from three references cost 23 credits.

Credits are reserved before the upstream request. If the provider request completely fails or no result can be stored, Imagine2 refunds the reserved amount. If a four-output request produces only three stored outputs, the missing output is refunded. Refund handling is idempotent, which means retries do not keep adding credits.

A successfully delivered image is billable even when you do not like it. Generative systems are probabilistic, and subjective dissatisfaction is different from a technical failure. Use low output counts while refining a brief.

Sign in and restore a draft

Google is the only V1 sign-in method. When you press Continue with Google on a public workspace, the browser saves prompt text, mode, frame, resolution, count, and selected local files in IndexedDB. After Google returns you to the original page, Imagine2 restores the draft and displays a restoration message.

Nothing generates automatically. Review the restored state and press Generate again. If the draft is older than 24 hours, it is not restored. A successful generation clears the pending draft.

New Google users receive 10 welcome credits valid for 30 days. The grant runs after the account is created and is recorded in the credit ledger.

Read the output darkroom

Results appear as a contact sheet. Hover or focus an image to see the selected model attribution and download control. The independent service statement remains visible below the result area.

Every output is first downloaded from the provider's temporary URL and persisted to private object storage. If one item in a multi-output response cannot be stored, Imagine2 keeps the successful stored items and refunds the missing portion. A temporary upstream URL is never treated as your permanent library.

Inspect results at full size. Look for malformed hands, inconsistent reflections, broken product geometry, accidental marks, misleading text, implausible edges, identity drift, and details that conflict with the brief. If the output represents a product or place, compare it against authoritative references.

Save, download, remix, edit, or delete

Unsaved inputs and outputs are retained for 24 hours. Press Save 30 days when a task is worth keeping. Saving extends the associated output expiration to 30 days.

History provides five useful actions:

• Save extends retention when the task is still temporary.
• Remix opens Studio in Create mode with the prompt prefilled, without carrying the original reference.
• Edit opens Studio in Edit mode with the selected output as a private reference.
• Download streams the owned asset with an attachment response.
• Delete removes the task and associated stored objects immediately.

Deleting is intended to be real, not cosmetic. Imagine2 requests object deletion before marking the database rows deleted. If storage deletion fails, the operation returns an error rather than pretending the asset is gone. Hourly expiration cleanup also leaves failed deletions eligible for retry.

Use iteration deliberately

When a result is close, identify the smallest meaningful change. Rewrite the prompt to describe that change and preserve everything else. Do not add a growing chain of contradictory adjectives.

For a composition problem, revise subject placement, camera, crop, or negative space. For a lighting problem, revise direction, softness, color, and practical sources. For a material problem, describe reflectivity, roughness, translucency, grain, and construction. For identity drift, choose a clearer reference and reduce competing transformations.

Keep a short record of what changed between attempts. A creative workflow becomes expensive when no one can explain why one request differed from the previous request.

Responsible publication checklist

Before publishing, ask:

1. Did I have the right to use every reference?
2. Did I obtain required consent or releases?
3. Could a reasonable viewer mistake this for authentic evidence or an unaltered photograph?
4. Does the context require synthetic-media disclosure?
5. Does the image make factual, medical, legal, financial, safety, or product claims that need independent verification?
6. Does it include recognizable brands, copyrighted characters, private people, or sensitive locations?
7. Is the alt text accurate and useful?
8. Have I checked the full-resolution file for artifacts and unwanted text?

The user remains responsible for publication decisions. Imagine2 is a generation workflow, not a rights-clearance service or professional reviewer.

Current V1 limits

Imagine2 shows the selected model directly in the workspace and on completed results. V1 does not expose region masks, transparent-background controls, Smart Resize, a complex canvas, or prompt enhancement. Controls stay limited to functions the selected model and current workflow can actually perform.

These limits keep the interface honest. A control should appear only when the selected model and product implementation can honor it. Model choices and available settings can change, so check the workspace before relying on a saved workflow.

A compact first workflow

The simplest answer to how to use Imagine Image 2.0 responsibly is to make every creative decision reviewable, even when the exact model or product surface changes.

Open Image Generator. Write a direction naming subject, frame, light, and exclusions. Choose 1K, one output, and the aspect ratio required by the final placement. Review the five-credit estimate. Continue with Google if needed, confirm the restored draft, and generate.

Inspect the result. If the concept is wrong, revise the prompt. If it is close, use Edit and name the specific change while preserving the successful parts. Save the useful task, download the final asset, add any required disclosure, and delete references you no longer need.

That is the Imagine2 workflow: direct, inspect, revise, and keep only what earns a place in the library.

Sources and further reading

• xAI image generation documentation
• xAI image editing documentation
• xAI multi-image editing documentation
• xAI acceptable use policy
• Imagine2 Acceptable Use Policy
• Imagine2 Privacy Policy