Describe the image you want to create
0/20000Advanced Settings
Tips for better results
- - Be specific about style, colors, and mood
- - Include lighting and composition details
- - Use artistic style references like "photorealistic" or "watercolor"
- - For image-to-image, upload multiple reference images for better results
- - Describe how you want to transform all uploaded images
- - If generation fails, try switching to a different model in the dropdown for different results and fewer restrictions
Nano Banana 2.1 for character and prop scene studies
Give a narrative scene a clear visual brief. On Soro2 AI, use Nano Banana 2.1 to explore characters, recurring objects and their placement, then compare each result against your own continuity notes.
Organize the evidence before describing the scene
A portrait, a costume study and a prop photo answer different questions. Tell the model what each reference contributes and which details the scene must retain.
Write the scene’s fixed details
A required prompt can hold up to 20,000 characters. State the character’s clothing, the condition of important props, the setting and the moment in the story, then separate fixed details from the change you want to explore.
Give each reference a role
Supply up to 14 JPEG, PNG or WebP images, with a 30 MB limit per image. Describe image 1 as the character reference, image 2 as the prop reference and image 3 as the location reference when that matches your inputs.
Frame the relationship that matters
Use the composition to show who is holding an object, where characters stand or what they notice. Select a ratio and 1K, 2K or 4K output for the level of review you intend to do.
A scene study, followed by a continuity check
Nano Banana 2.1 can turn a scene description and visual references into a still image for discussion. It does not manage a character bible or certify continuity. Keep your own record of wardrobe, prop condition, handedness and spatial relationships, and compare the result against it.
The 2.1 model is a separate choice from Nano Banana 2 on Soro2 AI. This page starts that model; fresh output settings are 1K, Auto and JPG. Choose PNG when needed and review each new image, even when you reuse the same references.
Define what must stay and what may change
For a second scene, distinguish the recurring coat and notebook from the new camera position or weather. Specific instructions give you concrete details to inspect afterward.
Keep a short discrepancy list
Record mismatched clothing, switched hands, changed object shapes or inconsistent positions. Use that list to revise the next prompt instead of assuming the references guarantee a match.
Make scene review specific enough to act on
Judge the generated image against a small set of story requirements, not only its overall appearance.
Know the cost of another scene study
Soro2 AI charges 2 credits for a 1K image, 4 for 2K and 6 for 4K. References do not add a separate charge. Choose the next revision because it addresses an identified issue.
Check relationships inside the frame
Verify which character is nearest the door, which hand carries the prop and whether the object is visible. These details can change the meaning of a scene even when the lighting looks suitable.
Keep images with your review notes
Download JPG or PNG and store the selected image beside your own scene notes. When comparing scenes, use the original references as well as the previous output; an earlier generated image may already contain a mismatch.
Scene prompts with explicit continuity checks
The requests below separate a visual intention from the checks a person should perform on the result.
Introduce a character with a recurring object
Upload a character portrait and a separate notebook photo; choose 3:2. Prompt: “Use image 1 for the character’s appearance and image 2 for the red cloth notebook with a black elastic band. Show the character seated at a train-station bench, gray coat buttoned, notebook closed in the left hand, soft overcast light, no added lettering.” Check the face, coat, holding hand and notebook closure.
Explore the next scene without losing the prop details
Reuse the relevant references and select 16:9. Prompt: “Use image 1 for the character and image 2 for the notebook. Place the character in a quiet archive reading room. Keep the gray coat buttoned. Put the closed red cloth notebook with its black elastic band on the table to the character’s right; both hands rest away from it. Medium-wide view, window light from camera left.” Compare the wardrobe and notebook shape with the references, and verify the requested placement manually.
Build a scene brief you can check afterward
The useful output is an image plus a clear decision about which details need revision.
Prepare a small reference index
Choose the relevant character, wardrobe, object and setting images. In your notes, assign each a role and list the story details that must be retained. Upload the selected references and write the prompt.
Generate one defined scene
Specify one visible moment, select the aspect ratio and set 1K, 2K or 4K with JPG or PNG. Read the credit cost, sign in and submit. New settings begin at 1K, Auto and JPG.
Check and record mismatches
Compare the completed image with the reference index and adjacent scenes. Note differences in identity, clothing, hands, props and placement; download the reviewed result or revise the brief. The task remains available in My Creations.
Questions about narrative reference work
Practical limits for reference inputs, output settings and keeping scene details under review.
How much does each scene image cost?
For Nano Banana 2.1 on Soro2 AI, a 1K scene image costs 2 credits, a 2K image costs 4, and a 4K image costs 6. There is no separate reference-image fee. Another generation is a new request with its own displayed cost.
How many character and prop references can I include?
Up to 14 images in total, not 14 per reference category. Each must be JPEG, PNG or WebP and no larger than 30 MB. Public HTTP or HTTPS image URLs must be accessible and meet the same constraints. Explain each image’s purpose in the prompt.
What composition ratios can I use for scene studies?
Choose from 15 options: auto, 1:1, 2:3, 3:2, 1:4, 4:1, 3:4, 4:3, 4:5, 5:4, 1:8, 8:1, 9:16, 16:9 and 21:9. Available output resolutions are 1K, 2K and 4K; choose the framing that shows the relationships you need to review.
Will recurring characters and props stay exactly the same?
References and detailed prompts can guide the request, but exact continuity is not guaranteed. Compare faces, wardrobe, object shape, color, handedness and placement yourself. Keep a separate continuity record and revise any result that conflicts with the story.
Can I save continuity instructions just by uploading references?
Write the instructions in the prompt for the request; both modes require nonempty text and allow up to 20,000 characters. Do not rely on uploads alone to explain what each image means. New settings are 1K, Auto and JPG, with PNG also available. This model has no Google Search setting.
What happens to credits if a scene request fails?
The generation credits are returned after the task is confirmed failed. A delayed response or a temporarily unavailable status query is not confirmation of failure. Revisit the existing task to check its state before requesting the scene again.
Bring one narrative scene into view
List the character and prop details that matter, assign reference roles and review the image with Nano Banana 2.1 on Soro2 AI.