I tried this with a short clip of a brass robot moving a terracotta flowerpot from a greenhouse floor to a workbench. The video offered some useful poses, along with footwork I would change. I picked eight moments, noted their times and put them on a sheet to refer to during later animation work.
Give the draft a clear physical task
I started with a new AI-generated greenhouse illustration. The robot stands on the left, a terracotta pot sits on the floor, and a wooden workbench is on the right. The full body, both hands, the pot and the bench edge are all visible, so there is room to see the lift happen.
I broke the action into four events: grasp the pot, lift it, set it on the bench and let go. That gave me something specific to look for in the result. The hands should keep hold during the lift, and the pot should be supported by the bench before they pull away. Looking only at the final pose could miss a problem in between.
I uploaded the reference to Image To Video on Kling 3.0, selected Standard, chose five seconds and disabled Generate Sound. Only the Start Frame was supplied; the End Frame was left empty. This was the complete submitted prompt:
Code: Select all
Use the uploaded greenhouse illustration as the start frame. Make one continuous five-second action with a completely locked camera and the same full-body framing. The small brass robot bends its knees and hips, reaches down with both gripper hands, firmly grasps the one terracotta flowerpot on the floor, lifts it as it straightens, then sets it down on the empty near edge of the wooden workbench on the right and releases it. Show the grasp, lift, bench contact and release clearly in that order. Keep the same single pot and seedling continuously visible; do not duplicate, teleport, melt or pass them through the robot or bench. Keep the robot feet planted, preserve its proportions and mechanical joints, and keep the greenhouse, bench and light stable. No walking, cuts, zoom, camera motion, text, extra robots or added objects. This is a motion draft for reviewing key poses and timing, not a finished rigged animation.
Sample the motion, then choose the poses
The downloaded draft was 1280×720 at 24 frames per second, with 121 decoded frames and a duration of about five seconds. I decoded it with FFmpeg through imageio-ffmpeg, saved every fourth native frame and recorded each source index before laying out the sheet. This produced 31 samples at six samples per second, without interpolating images or correcting poses.
Laying out regular samples made it easier to spot where the action changed: the robot bending down, its hands closing around the pot, the pot leaving the floor and the hands withdrawing after placement. I chose eight of those moments for the sheet below. The pictures are untouched frames from the downloaded video.
In the samples I checked, the grasp, lift, carry and placement followed a readable order, and the pot stayed recognizable. That sequence was worth keeping as a reference. The feet needed a separate look.
There are still frames between the samples. Before using the motion, play the clip at its normal speed and look more closely around the lift and release to check how those moments connect.
Keep source time separate from animation time
For each chosen pose, I noted its frame number in the original video and its time in seconds. That makes it easy to find the same moment again. This file runs at 24fps, with source frames numbered from zero, so frame 48 is at 2.000 seconds: 48 divided by 24.
If the later Moho animation also runs at 24fps and starts at frame 1, the corresponding planning number is one higher. I put the two sets of numbers side by side:
| Source time | Source frame, from 0 | Selected event | Planned frame, from 1 |
|---|---:|---|---:|
| 0.000s | 0 | Starting stance | 1 |
| 1.000s | 24 | Reach and compression | 25 |
| 1.500s | 36 | Two-hand grasp | 37 |
| 2.000s | 48 | Lift begins | 49 |
| 2.500s | 60 | Upright carry | 61 |
| 3.500s | 84 | Reach toward bench | 85 |
| 4.333s | 104 | Bench contact | 105 |
| 4.833s | 116 | Release and recovery | 117 |
The chosen grasp and lift poses are half a second apart, as are bench contact and release. Those intervals give me a starting point for the timing. I can then adjust them to suit the character's sense of weight and make each action easier to read.
The right-hand column is a plan; no keyframes have been inserted in Moho. For another frame rate or starting frame, recalculate from the seconds column. Reusing the frame numbers without that adjustment could make the action faster or slower.
Mark the movement you will change
The clearest departure from the prompt was in the feet. Although I asked them to stay planted, the robot adjusted its stance before grasping the pot and stepped toward the bench. Those movements need their own decision before I borrow the poses.
I would keep the body compression, the two-handed hold and the order of setting the pot down before letting go. I would rethink the stance and placement path, rather than carry the unwanted steps into the animation.
A short note beside each pose helps: “Keep the arm silhouette; redesign the foot contact” says more than “copy frame 84.” For placement, I would add “Support the pot on the bench before the hands withdraw.” Those notes keep the important decisions visible when it is time to build the contacts in the animation.
Use the sheet to plan your Moho shot
The sheet and timing list are now ready to use when planning the shot. With your own character in Moho, start with the grasp, lift, placement and release poses, then work on the spacing between them. If the grasp feels rushed, give the hands more time to settle around the pot. The generated clip's pacing is a starting point you can change.
Keep the original source time in the notes even after moving a pose to a different animation frame. You can then see both where the idea came from and where you chose to place it, which makes later revisions easier to compare.
Before moving from the sheet to animation, check:
[*]Each chosen image has a source timestamp and frame index.
[*]Planned timeline numbers use the intended frame rate and starting convention.
[*]Grasp, lift, placement and release are separate events where the action needs them.
[*]Unwanted steps and uncertain contacts have explicit revision notes.
[*]Full-speed playback supports any timing you decide to retain.
Friend, the Moho official stated a long time ago that they would not introduce AI into the Moho system. Your research is great, but you may not receive a response. I think it's more like a game (turning every aspect into standardized modules) than an animation.