How Many Ways Are There to Generate AI Video? When to Use Multi-reference vs First/Last Frame
An overview of common AI video generation methods, comparing Multi-reference and First/Last Frame use cases, with a workflow for choosing paths, preparing inputs, and reviewing results in NadouPro.
How Many Ways Are There to Generate AI Video? When to Use Multi-reference vs First/Last Frame
This article explains common AI video generation methods, compares the use cases for Multi-reference and First/Last Frame, and shows how to choose a path by shot goal, prepare inputs, and review results in NadouPro.
How many ways are there to generate AI video?
For creators, AI video generation is usually divided first by “what the input is” into two types: Text to Video and Image to Video. Text to Video relies only on written descriptions to create images, making it suitable for exploring abstract scenes. Image to Video uses uploaded images as visual anchors, making the results more predictable. If beginners already have character or scene images, the Image to Video path is usually the better first choice.
In NadouPro's Video node, you can also choose further by “control method”: Text to Video, First/Last Frame, Multi-reference, and Motion Control. Multi-reference and First/Last Frame are the two most often compared: the former is good at blending multiple features in one generation, while the latter is good at locking the start and end of a shot and its motion destination. You can first use an Image node on the canvas to finalize the look, then connect it to a Video node for generation, without repeatedly exporting and importing throughout the workflow.
What are the boundaries between Text to Video, Image to Video, and reference-based control?
Text to Video only takes a prompt as input, offering high freedom, but the subject can drift easily. Image to Video lets the image define “what it looks like,” so the prompt should focus more on action, camera movement, and atmosphere. Reference-based video generation is not a separate mysterious third entry point; it strengthens control on top of Image to Video by using multiple references or keyframes to constrain identity, style, and the starting and ending images.
You can remember it this way:
- You only have an idea and no finalized image yet: use Text to Video to test a direction
- You already have a static image you like: use Image to Video to make it move
- You need to lock character, product, and style at the same time: use Multi-reference
- You need to define where the shot starts and where it lands: use First/Last Frame
What scenarios are Multi-reference and First/Last Frame suited for?
Multi-reference: blending multiple features at once
The core of Multi-reference is feature blending. You can inject a character face, product appearance, scene style, and even reference camera movement or audio into the same generation. It is suitable for maintaining character consistency across shots in short dramas, keeping both the Logo and spokesperson intact in product ads, and producing shots where “character setting + scene + style” must land in the same frame.
When preparing inputs, keep them few and precise: clear subjects, clean backgrounds, and similar styles. The prompt should clearly state the purpose of each reference, such as “reference image 1 locks the product appearance, reference image 2 locks the character identity, and the prompt describes handheld follow movement and the final framing.” The maximum number of references supported varies by model, so follow the description of the model you select in NadouPro.
First/Last Frame: locking the start, end, and motion path
In First/Last Frame mode, the first frame defines the opening, the last frame defines the destination, and the model fills in the transition between them. It is suitable for connecting storyboard frames, aligning a clip with existing footage, guiding a scene toward a specific visual ending, and transitions “from state A to state B” such as transformation, disintegration, or form switching.
The prompt does not need to repeat the appearance and scene already present in the frames. Focus on the dynamic process, shot size changes, and camera movement instead. For example: dolly in from a Medium Shot, fast orbit, slow-motion sword withdrawal ending in a freeze frame. If you provide only the last frame, the model will decide how to get there on its own, giving you weaker control. For key shots, it is recommended to provide both the first and last frames.
How should you choose and organize an executable workflow in NadouPro?
Choose based on the goal of the shot instead of getting stuck on the model list first:
- Clarify what this shot needs to lock: identity, product, starting and ending images, or just atmosphere exploration.
- Prepare inputs: character or scene look-development images, a Multi-reference asset pack, or key first and last frames.
- Select the corresponding video mode on the NadouPro canvas, and write a prompt that clearly states the subject, action, scene, and shot.
- First do a test generation at a lower resolution, watch the full segment, and check whether the camera movement and subject are stable.
- Change only one variable at a time, such as the prompt, reference, or frame. After you are satisfied, increase the resolution for the final output.
Stage outputs can be: look-development images or keyframes → test clips → shot materials that pass review. When checking, do not look only at the first frame. Focus on whether the middle section breaks down, whether the character's face changes, and whether the final framing reaches the intent of the last frame.
Suitable and unsuitable cases
Multi-reference is suitable when “multiple elements must be correct at the same time.” It is not suitable when references conflict with each other or when the prompt conflicts with reference instructions. First/Last Frame is suitable for “transitions with a clear path.” If the start and end are too different and there is not enough motion description, the middle may show stiff jumps. Pure Text to Video is better suited for concept exploration and is not suitable for strictly locking faces in serialized episodes.
NadouPro can cover path selection, generation, and shot organization, but the final pacing, dialogue refinement, and final edit still need creator oversight. Capabilities depend on the current platform and the selected model.
Related Questions
What is the difference between Text to Video, Image to Video, and reference-based video generation?
Text to Video relies only on text, offering high freedom but lower stability. Image to Video uses an image as an anchor, making it more predictable. Reference-based video generation strengthens control on top of Image to Video, using Multi-reference or First/Last Frame to constrain identity, style, and the start and end. If you have a finalized image, prioritize Image to Video or reference-based control.
How should I prepare reference assets for more stable Multi-reference generation?
Choose a small number of references with clear subjects, clean backgrounds, and similar styles, and specify each one's purpose in the prompt. Provide characters, products, and scenes separately whenever possible to avoid overloading one image with too much information. Start with a lower-resolution test generation, then improve quality after confirming the direction of feature blending.
What shots are suitable for First/Last Frame, and how should I check whether the result is usable?
It is suitable for connecting storyboard shots, aligning with existing footage, guiding the clip toward a specific final framing, and morphing transitions from state A to state B. After generation, watch the full segment: check whether the middle is smooth, whether the subject drifts, and whether it truly reaches the intent of the last frame. If it deviates, simplify the prompt and let the two frames carry more of the direction.
How do I choose a generation method by shot goal and organize the workflow in NadouPro?
First decide whether the shot needs to lock identity, multiple features, or the start-to-end path, then choose Text to Video, Multi-reference, or First/Last Frame in the Video node. You can use an Image node to create a look-development image, connect it to a Video node for a test generation, iterate on the canvas, then increase the resolution and move into editing after approval.
Summary and next step
AI video generation can first be understood as Text to Video or Image to Video, then executed in NadouPro through control methods such as Text to Video, First/Last Frame, and Multi-reference. Multi-reference is strong at placing multiple features in the same frame, while First/Last Frame is strong at locking the motion path. Next step: choose one real shot, prepare the matching inputs, complete a lower-resolution test generation in NadouPro, and review the full segment.
Contains AI-generated information that may be incorrect. Please verify independently.