Nano Banana Stills After the Sora API Sunset: Seedance, Kling, and Veo

OpenAI already discontinued the Sora app. The Sora API dies on 24 September 2026, and OpenAI’s own deprecation table names no replacement. If your loop was Nano Banana stills → Sora clip, the stills half still works. The video half has to leave OpenAI.

This is a short companion for that handoff: approve Gemini stills, then generate one non-Sora beat on Seedance, Kling, or Veo. It is not the wedding invitation and Seedance-style teaser workflow. It is not the Seedance 2.5 bullet-time / product-drop guide. Do not paste those couple or coffee-orbit prompts here.

This page is unofficial. OpenAI, Google, ByteDance, and Kuaishou did not review it.

What OpenAI actually shut down

Two clocks, both from OpenAI:

1. Consumer Sora. The Sora discontinuation help article states that the Sora web and app experiences were discontinued on April 26, 2026, and that the Sora API will be discontinued on September 24, 2026.
2. Developer Videos API. The OpenAI deprecations page records a 24 March 2026 notice that the Videos API and the Sora 2 aliases/snapshots (sora-2, sora-2-pro, sora-2-2025-10-06, sora-2-2025-12-08, sora-2-pro-2025-10-06) are removed on 24 September 2026. Every row in that block shows recommended replacement: ---.

There is no official OpenAI video successor on that table. Do not wait for a ChatGPT-shaped Videos API. Export anything you still need from Sora while export exists, then treat Sora prompts as source notes, not paste-ready jobs.

What this companion is for

Use this page when you already have (or can make) Nano Banana stills—portraits, locations, wardrobe plates, storyboard frames—and you used to send those files to Sora as image-to-video.

Do not use this page when:

• You need print-ready invitations, save-the-dates, or couple-identity teasers → wedding stills to Seedance-style clips.
• You need pack shots, coffee spills, CCTV orbits, or product drops → Seedance 2.5 bullet time from Nano Banana product stills.

The job here is narrower: keep the stills lock, change the video vendor.

Why the stills still matter

Nano Banana is the common name for Google Gemini image models. Those models were never Sora. They still generate and edit stills: identity, wardrobe, architecture, lighting, readable type.

Sora (and every replacement in this class) is better at one short motion beat than at inventing a person, a room, or a logo from a paragraph. Mixing both jobs in one prompt is how faces drift and rooms get swapped.

Lock four things in stills before any clip:

1. Subject. Face, silhouette, or hero object—whatever must not mutate.
2. Wardrobe or surface. Clothes, materials, colorway.
3. Place and light. Time of day, architecture, key direction.
4. The beat. One action. One camera path. One duration.

Those stills do double duty. A portrait can stay a still. The same file becomes the identity reference for an 6–12 second walk, glance, or room push-in.

New to the image model? Start with the Nano Banana AI image generator guide. For copy-paste stills language, use the prompt guide. When a face must survive a new pose, use the character consistency guide.

Stills → Seedance, Kling, or Veo (unofficial)

Pick one video model per beat. These are independent products, not a ranked “Sora killer” list, and none of them is an official OpenAI replacement.

Lane	Typical reason people reach for it after Sora	What the stills must lock
Seedance (ByteDance Seed)	Short reference-to-video with joint audio-video on some surfaces	Subject + place; say what each @image / Image N must not control
Kling (Kuaishou)	Image-to-video with strong camera-path language	Start frame (and end frame if the UI takes one); name the move
Veo (Google)	Gemini-adjacent video when you already live in Google’s stack	Same identity stills; do not assume Sora field names map 1:1

Limits, duration, audio, and which app exposes which model change. Recheck the vendor’s current page before you quote a number.

1. Approve 2–6 stills

Fewer files with one job each beat dumping a Sora gallery into a new model. A compact generic set:

#	Still	Job in video
1	Hero identity or hero object	Must not change
2	Three-quarter or opposite angle	Stops the orbit from inventing a new silhouette
3	Wardrobe / material close-up	Clothes or surface lock
4	Location or lighting plate	Architecture and time of day; ignore stray faces
5	Optional end-frame still	Only if the tool accepts a last frame
6	Optional motion plate	Path and pacing only—say what not to copy

Keep people out unless you have consent.

2. Assign one role per file

Image 1 locks identity (face / silhouette). Do not take its background.
Image 2 locks the same subject from a second angle.
Image 3 locks wardrobe or materials only.
Image 4 locks location and lighting. Ignore any people in this file.

If the UI uses @image1 (Seedance-style) or “first frame / last frame” (Kling-style), keep the same sentences. The syntax changes; the rule does not.

3. Rewrite the Sora brief — do not paste it

Sora prompts often smuggle duration, cameo names, and UI-only flags. Other models will ignore or misread those. Keep:

• format and duration
• subject + one action
• reference roles
• a timestamped timeline
• camera path
• continuity holds
• audio only if the model writes audio
• hard constraints (no extra people, no invented type, no second location)

Drop Sora-only product language. Then generate one beat. Review the stills lock before you extend or stitch.

Unofficial skeleton (neither wedding nor product orbit)

FORMAT: 16:9, 8 seconds, photoreal, no text overlay.

SUBJECT + ACTION: The approved person from the stills walks
the approved street at the approved time of day, then the
camera settles on a medium hold.

REFERENCE ROLES:
Image 1 locks identity. Image 3 locks wardrobe.
Image 4 locks the street and light. Do not invent a second city.

TIMELINE:
0–2s: medium-wide, subject enters from camera left.
2–6s: continuous walk, stable tracking, no crash zoom.
6–8s: settle, face readable, hold.

CAMERA: gentle track, eye-level, 35–50mm feel. Not FPV. Not a drone dive.

CONTINUITY: same face, same outfit, same block. No extra principals.

AUDIO: street room tone only if the model writes audio. No score. No subtitles.

CONSTRAINTS: no beauty-filter identity change, no new wardrobe, no signage you cannot spell.

That is a generic stills-to-clip beat: walk, glance, or room push-in. For vows and invitation type, use the wedding guide. For frozen coffee and CCTV pack orbits, use the bullet-time guide.

Sora prompt remap (H3 Max)

This site owns the Nano Banana stills half. Field-by-field remapping of a Sora-style video prompt onto Seedance, Kling, or Veo lives on H3 Max:

• Intended remap page: Sora → Seedance / Kling / Veo prompt remap
• Hub fallback if that URL is not live yet: h3maxvideoprompt.com

This site does not sell that builder. Use it only if it saves you from rewriting duration, camera, and reference roles by hand.

Related Nano Banana guides

Stills half, on this site:

• What Nano Banana is and how to generate or edit images
• Copy-ready Nano Banana prompt guide
• Keep the same faces across new poses
• Portrait lighting and identity prompts
• Browse the Nano Banana guide library

Different stills-to-video jobs (do not mix prompts):

• Wedding invitation posters → Seedance-style clips
• Seedance 2.5 bullet time from Nano Banana product stills

Disclaimer

This is an unofficial creator note. Sora and the Videos API are OpenAI products; dates and the blank replacement column come from OpenAI’s help and deprecations pages and can change—recheck them. Nano Banana is a common name for Google Gemini image models. Seedance is a ByteDance Seed video model. Kling is a Kuaishou video model. Veo is Google’s video model. None of those companies reviewed this page.

Do not upload other people’s faces without consent. Short AI clips can support mood boards, social tests, and previz. They do not replace a contracted shoot, rights-cleared likeness, or platform rules that require real footage.