Overview
Video generation in Pencil can be used not only to animate scenes and subjects and apply transitions, but also to create smooth transformative visual effects - often referred to as 'morphs'. These effects involve one image or concept transforming into another, and can be particularly powerful for stylised, surreal, metaphorical or experimental content.
Using Pencil's integration with the AI model Runway you can really increase your chances of a successful morph by using both a start frame and and end frame. A good choice of images for your start and end frames, combined with a strong, descriptive prompt that encourages continuity and fluidity is an excellent starting point.
It is important to note that morphing is an experimental or emergent use of video generation at the moment. This means it is not guaranteed, not always perfectly reliable, but possible given the right conditions.
Using start and end frames to guide morphs
Combined with the right prompt, start and end frames can be very useful when creating morphs.
With Runway, you can upload both a start frame and an end frame
Combined with the prompt, the model will attempt to interpolate between the start and end frames, generating a sequence that smoothly transitions from one to the other
This will be most successful when used with visually related or compatible subjects - e.g. face to face, object to object, or stylised transformations
Both the start and end frame choices should be visually distinct but not wildly dissimilar in structure, background etc.
Writing effective morph prompts
Use verbs and adverbs that describe transformation: e.g. morphs into, melts, evolves, dissolves into, morphs seamlessly into...
Include terms that guide pace: e.g. gradually, seamlessly, over time, slowly...
Include contextual continuity: e.g. 'As the moon rises, the eagle morphs into an owl' may perform better than 'the eagle turns into an owl'
Limitations and tips
Morphing yields the most success when the model can interpolate clear intermediate states. Very complex or conceptually distant transitions (e.g. 'a bed becomes a volcano') may yield chaotic or bizarre, and ultimately unwanted results.
Try to keep background elements simple or at least consistent across start and end frames to reduce visual noise that could interfere with the model's interpretation of the desired outcome.
Review outputs closely - looking out for any artefacts or unnatural movements in both the foreground and background.
Consider using seeds in image generation to create your start and end frames.
Example
The two images below were generated in Pencil, the latter using the seed from the former, with the prompt changed to generate a kitten with wings.
They were then used as start and end frames with a simple prompt along the lines of 'kitten playfully running along a path as wings magically grow from its back.'
The model was able to effectively create a morph effect where the kitten continues running and wings appear to grow from its back.
