Tough Cluck

Blog

  • Tough Cluck Tutorial

    Anyone can do this. My workflow to make Tough Cluck was as follows.

    Step 1. Outline your script.

    I didn’t have a finished script when I started, but I needed a rough outline to know what characters and locations I’d be working with. The more crucial the character or location, the more you want to invest in writing a descriptive, reusable prompt.

    Step 2. Set the scene(s) for prompt consistency.

    As a preamble I included consistent context & character descriptions in most prompts, i.e. “Tough Cluck is an ex cop from a gritty 1970s cop movie set in San Francisco. Cluck just happens to wear a full body chicken suit at all times. Include sound effects but do not include music. Do not include subtitles.”

    I also took some shortcuts to aide consistency. A guy in a chicken suit, or bad guys wearing ski masks, will look more consistent over multiple generations than a clearly visible face. I left the main villain in the dark so I could reuse them more easily in the next project.

    Even my choice for Cluck’s car was intentional, choosing something iconic like a VW bus which has a ton of training data, is going to give you more consistent output over multiple prompts. Understand and leverage the limitations of the AI.

    Step 3. Generate (and regenerate) clips

    Start generating. I used Gemini veo3 for most of this project but other creators are leveraging a host of different tools, and the capabilities change weekly. If you are starting completely from scratch veo3 is a helpful “all in one” to start getting your hands dirty, figuring out how to tune your prompts, but if you get more serious you’ll find yourself wanting to incorporate more tools. Search reddit for the latest and greatest tool recommendations.

    Part of the process is trial by error as you start out. If you can’t tell I set a pretty low quality bar for this project. I probably took the “first take” for 40% of the final scenes. 1 out of 3 generations for another 50%, and for about 10% of scenes I went up to maybe 5-8 attempts or needed to do additional post processing.

    Veo3 could not handle complexity. It could really only handle a single major action & dialogue within each clip, maaaybe two. For instance, “Cluck <does a thing> and says ‘<dialogue>’” tended to generate fairly reliable output. “Cluck <does a thing>, says ‘<dialogue>’, then Cluck <does another thing>” was almost always a hilarious failure. I often found I had to split what I thought could have been a single clip into two separate, more basic prompts and clips.

    Step 4. Additional audio and misc AI generation

    Some clips had no dialogue. Others had subtitles (even though the prompts said “Do not include subtitles”). I used ElevenLabs to generate additional dialogue audio, and even recorded a few clips of my own, i.e. the clip of high school Cluck getting turned down by giggling girls. My kids loved being included.

    For background music I used Suno. Suno can work with custom lyrics or auto generate its own for you based on a prompt. Personally I like to work on lyrics with help from a chat bot (ChatGPT in this case), adjust, and then drop those in to Suno as custom lyrics.

    As I neared completion of all the clips I also started generating some of my “marketing materials” for fun, like profile and banner images for a TC YouTube profile, but that’s all pretty optional.

    Step 5. Movie editing

    I used Adobe Premiere Pro (free trial) to start stitching everything together. Many people recommend DaVinci as an affordable alternative. I’ll try it for my next project.

    Premiere is powerful but there was a definite learning curve. There are many tutorials out there if you want to start from scratch. I had enough very rusty Adobe Photoshop and other amateur editing experience from hobby photo & video editing that I just dove in, leaning heavily on ChatGPT for tailored tutorials to guide me when I knew what I wanted to do but not how to do it in the software.

    Step 6. Post & share!

    Post to YouTube or share directly with your friends and family!

    Step 7. Lessons learned & how to do better

    My approach was low effort and scrappy! There are creators out there using similar tools to generate much more polished output. Check out MetaPuppet, itsprobablyai, and many others. I follow /chatgpt and /aivideo on reddit for inspiration. Some tips and tricks out there right now that I may incorporate in my next project:

    • For better character consistency, take a still shot from the end of the clip (or use a single core shot of a key character every time) and upload it to your video generator as part of the prompt & context. Downsides are that some tools won’t create dialogue from it, as a deep fake prevention.
    • If you want more control over your characters, you can go a completely different path with a green screen & tools to map a gen AI face or character to your own, so you can act out the scene yourself. You can couple this with ElevenLabs to alter your voice to a custom character as well.
    • Get creative, mix and match different tools! One example: MetaPuppet took an 8 second audio/music clip from veo3, fed it to Suno as a theme source, and extended that to generate soundtrack music for a longer video.

    Have fun creating!

  • Introducing Tough Cluck

    Introducing Tough Cluck

    I made a terrible movie…

    When veo3 dropped, I was intrigued. How far could I take the 8 second clips it generates? And so began the rabbit hole adventure that is Tough Cluck.

    This was a proof of concept for myself and my kids, to get them thinking creatively about gen AI capabilities and limitations. We had a lot of fun. We know this is terrible and we had a blast making it.

    But why?

    Why make a series of shorts about a half-man, half-chicken hero?

    Because it’s absurd. Tough Cluck is, on the surface, a playful homage to the films I grew up with. Gritty 70s cop dramas, over-the-top 80s action flicks, and stylized 90s thrillers. It also highlight serious capabilities and limitations of AI.

    AI right now is both incredible and deeply flawed.

    On the one hand, it opens doors for everyday creators. I can spin up a fake action hero on my phone. Twenty years ago, that would have taken a team, expensive software, and a lot more skill.

    On the other hand, the results are limited, biased, and often glitchy.

    • veo3 wouldn’t generate audio when kids were in the scene. That’s probably a very good thing – I’m not complaining – but it forced us to record some of our own audio or explore other workarounds.
    • Tons of errors. In one scene a prompt that mentioned “seals barking” resulted in seals barking like dogs. The “magic” still requires a lot of trial and error, curation, and polish. I could have fixed a lot of these scenes with more prompts & retries, but I like highlighting the AI fails in this context. Tough Cluck 2 will get a little more serious and polished.
    • Bias is rampant. Without specifying anything in my prompts, the majority of scenes defaulted to white characters. Is that the fault of the training data, the model, or both?

    There’s also the fascinating question: who owns this?

    • The AI model spits out a scene based on the countless works it was trained on. Is the result mine? Yours? Everyone’s?
    • I used commercial AI tool licenses (vs free accounts) throughout the process so I ostensibly have more ownership and rights over the output, the sum of the parts, but who owns the parts is still a legally evolving question.

    I do know that these tools democratize creation in a way that’s exciting. I can make (bad) movies now that I couldn’t before. But they also risk displacing the very professionals who inspired me in the first place.

    Tough Cluck is both a joke and a statement. A celebration and a critique. It’s me playing with the tools at hand, while openly wrestling with the bigger implications. I hope you hate or enjoy it. Even better if you feel conflicted.

    (and yes, AI helped me workshop this draft).