Codex Image Generation + Seedance Image-to-Video: Celestial Palace Chinese-Style Cinematic Tutorial (2026)
This tutorial uses Codex to generate a Chinese-style "Celestial Palace" image, then uses Jimeng (Seedance 2.5) to turn it into a dynamic cinematic clip of an ancient bridge over a sea of clouds with immortals walking slowly along it. You don't need to understand how AI works — just follow the steps, paste the prompts, and you'll get your clip in about 15 minutes.
Bottom line first: three steps to a finished video — ① Codex generates the image → ② Codex writes the video prompt based on the image → ③ Jimeng (Seedance 2.5) turns the image into video. Three things to know in advance: ① keep the figures very small (about 2% of the frame), with plenty of sky and clouds so the shot has room to "move"; ② the video prompt should be based on your own image, not copied from someone else; ③ Codex has a daily quota — if image generation fails, first check whether the quota is used up.
Contents:
- Prepare Your Tools
- Step 1: Generate the Image with Codex
- Step 2: Generate the Image-to-Video Prompt
- Step 3: Generate the Video with Jimeng
- Pitfalls to Watch Out For
- FAQ
- The Final Result
Prepare Your Tools
- Codex (OpenAI): generates the image, then writes an image-to-video prompt based on it.
- Jimeng (Seedance 2.5): image-to-video, turning the image into a dynamic video.
- The three prompts provided in this article (see below).
Step 1: Generate the Image with Codex
Paste the following prompt into Codex to generate the celestial palace image:
A majestic ancient stone bridge stretches from the foreground into the distance, curving in a grand arc toward a hazy amber horizon.
The deck follows the arc — clean, wide, and uncluttered. Polished pearl-gray stone and lacquered surfaces are complemented by massive stone piers.
In the left foreground, a huge carved column of pearl-gray stone dominates the composition; more towering pearl-gray stone columns recede along the bridge.
On the right: an open, boundless sea of clouds. Two Chinese immortal female figures — one in a bright red robe, one in a white robe — walk slowly along the curved bridge, tiny against the grand architecture. The figures occupy about 2% of the frame height, like a small fleck of color, walking slowly forward. Their bodies stay still; fabrics and long hair catch the same high-altitude breeze and trail backward. Empty space around them spans more than ten times their height.
Warm afternoon side light creates a soft atmospheric haze.
In the distance: grand Chinese temple architecture faintly visible among drifting clouds; the mood is serene and solemn, like a Chinese immortal realm.
Epic Eastern aesthetics — majestic yet ethereal, with striking proportional contrast.Expected result: a composition with the ancient bridge arcing from the foreground into the distance, seas of clouds around it, and distant celestial palaces faintly visible — the two immortals take up only about 2% of the frame.
No prompt on hand? You can also drop a similar Chinese-style ancient bridge image into Codex and ask it to generate something based on it:
Step 2: Generate the Image-to-Video Prompt
Put the image you just generated back into Codex, paste this instruction, and let Codex look at the image and write the video prompt:
Describe how this image would look in motion. Give me image-to-video prompts.Expected result: Codex will return a combined prompt covering camera movement, environment motion, and character motion. Use its output as the source of truth — it doesn't have to match this article's example word for word.
Step 3: Generate the Video with Jimeng
Put the image and the prompt from the previous step into Jimeng, choose image-to-video mode, and paste the prompt:
You can use this recommended motion-direction prompt directly, or the one Codex generated in Step 2:
Recommended motion direction. Camera: slow dolly-in while rising slightly, making the bridge's depth and the sea of clouds more breathtaking. Environment: the clouds roll and drift slowly, like immortal mist encircling the scene. The distant celestial palace is veiled in thin mist, clouds gliding slowly past the architecture. Golden sunset light shifts, rays breaking through the clouds. The bridge deck faintly reflects the changing sky. Characters: the two immortals walk slowly forward along the bridge. The red-robed woman's hem and long hair flutter gently in the high-altitude breeze. The white-robed woman's sleeves lift softly. Keep the figures very small in frame to emphasize the vastness of heaven and earth.Expected result: after generating, check three things — whether the clouds flow naturally, whether the figures stay very small, and whether the golden evening light breaks through the clouds. If you're not satisfied, fine-tune the prompt and regenerate.
Pitfalls to Watch Out For
- Codex quota: image generation has a daily ceiling. If it fails, first confirm your quota — don't retry over and over.
- The source image decides everything: if the subject is too large or the composition too packed, the video looks "stiff." When generating the image, be sure to leave large open areas of sky and clouds.
- The prompt must follow the image: every image has a different composition and lighting, so base the video prompt on your own image — don't force someone else's.
- Jimeng queueing: video generation usually takes a little waiting. Be patient — a clip almost always comes out.
FAQ
I don't have a prompt, but I still want to make a celestial palace video. What do I do?
Just use the prompt in Step 1 of this article and let Codex generate an image of an ancient bridge over a sea of clouds with immortals beside it. If you're not happy with the generated image, you can drop a similar Chinese-style ancient bridge image into Codex and ask it to generate or rewrite one by reference.
Why does my Jimeng video look stiff and not move?
Nine times out of ten it's the source image: the subject takes up too much of the frame, the composition is too packed, or there's no large area of sky or clouds. When generating the image, keep the subject very small and leave plenty of empty scenery — that's what gives the video room to flow naturally.
Codex generated the image but immediately says my quota is insufficient. What do I do?
Codex has a daily quota, and image generation burns through it faster than chat. Wait for the quota to reset the next day, or check the remaining credits on your current plan. Don't keep retrying while the quota is exhausted.
The Final Result
Here's what the finished clip looks like using this workflow:
