SSH Portfolio | R&D My Journey with Generative AI in Media Production

My Journey with Generative AI in the creative industries

Hey there, fellow AI geeks! I've been diving into AI for media production since discovering Synthesia and Ebsyth in 2018. Shortly after, I started using it in my daily creative processes when Adobe rolled out Neural Filters in 2019.

As a Creative Director and technologist who has led teams at Sage, Qredo, Thrive (an AI-driven learning platform), and UNLEASH.ai, I use these tools to spark creativity and streamline workflows, delivering award-winning campaigns like the 2020 Telly Awards. Here's how I use them, plus some ethical insights and tools worth checking out.

Content Creation and Research

Grok: Currently, my top pick over ChatGPT, DeepSeek, or Gemini...subject to change, being based in the EU, we are not receiving the latest AIs when released due to restrictions imposed in my region. I use it to analyse video transcripts generated in Adobe Premiere, which speeds up my text-based editing back in Premiere. It's a brilliant shortcut, but the final content always needs a human touch to feel alive. Grok also helps with copywriting, content ideation, and research. Ethical Rating: 3/5 - xAI is transparent about its mission, but some X users raise concerns about data privacy and bias risks.

Perplexity AI: A fantastic research tool with source citations, perfect for digging into media trends. Ethical Rating: 3/5 - It prioritises transparency, but details on data sourcing are limited.

Audio and Music Production

Adobe Gen-AI: Handles automatic music editing and voice-over enhancements for crisp audio. Ethical Rating: 4/5 - Trained on licensed Adobe Stock and public domain content. I prefer using AI tools over AI generators for commercial work and supporting other creatives' jobs.
Adobe asked me not to show their logo, which is a shame, as they have the best review on this page and have been assigned the best ethical scores.

ElevenLabs, Artlist, MotionArray: I confess, I have used all three for AI-driven voiceovers and voice cloning, delivering realistic results. Typically, when there isn't the budget or the time to work with a human VO artist/service, what is worrying is that some of the AI-generated voices are more authentic and human than some of the actual voice artists I have hired before, whose delivery was too polished and robotic.

My preferred approach is NOT text-to-voice but voice-to-voice. I will act out the script myself, using my desired tempo, dramatic pauses, and emphasis, and use this to direct the AI-generated version. Hopefully, subtle changes in the scratch VO, such as volume and emotional nuance, will also be translatable in the future.

Ethical Rating: 2/5 (ElevenLabs), 3/5 (Artlist, MotionArray) - ElevenLabs lacks clear training data details, while Artlist and MotionArray use licensed content but aren't fully transparent about AI processes. Conversely, ElevenLabs has been very open and ethical when it comes to working with music studios for their new AI-music generator

Soundraw: Creates royalty-free music tailored to video moods, great for quick edits. Ethical Rating: 3/5 - Uses licensed data, but training details are vague.
As mentioned before, I still prefer to find the track I want from a real artist, but music editing other people's tracks to fit my needs can be a pain (unless you are purchasing an entire music kit like the Premium Beats platform). Using an AI to select where in the track the mood should change, and quickly editing the duration, on the other hand, is a game-changer and very welcome.

Video Editing and Enhancement

Adobe Gen-AI: Crucial for precise retouching and editing of footage and images. Out of all of the latest AI products that have emerged in recent years, these are the ones I use the most, and I mostly use these tools in Premiere, Photoshop, and After Effects...less so in Illustrator. Ethical Rating: 4/5 - Strong on licensed data, but the Midjourney image issue slightly clouds its ethical stance. As a keen follower of Adobe Max, the sneak peeks always pique my interest. Watch this space! Adobe asked me not to show their logo, which is a shame, as they have the best review on this page and have been assigned the best ethical scores.

Veed and Descript: It is one of my missions to level-up non-creatives. For a pro video editor or motion designer, they will likely want to stick with their pro tools, but for EVERYONE else this is an assume tool. Having said that, I have used Veed.io's AI-eye-fixing tool when the subject's eyes can be seen reading a script. I would love to have something similar for camera A/B interviews, to remove those moments when the interviewee's eyes flick to one of the cameras, breaking the 4th wall. And instead have the eyes locked off screen towards the interviewer. Veed 'Fabric' looks like it has a lot potential as a voice+video generator, including lip sync. Veed.io uses content uploaded by its users (with opt out option for paid users), so it is probably a 4/5 on th ethical scale, but I cannot find any information on Frabic yet, so until then, a provisional 3/5 is a safe bet in the short term.

Topaz Video AI, AI Scale-Up, Depth Scanner, Speedx AI Plugins: These boost video upscaling, depth mapping, and motion enhancements. Ethical Rating: 3/5 - Focused on enhancement, but training data sources aren't well-documented. Of course, Runway can also do all of the above, but if you are like me, I am SICK TO DEATH of having to pay for yet another subscription. I would much rather just pay for a plugin I can use locally.

Visual Editing and Enhancement

Synthesia: Ideal for quick video avatars and personalised content. I love Syntheia, ever since pitching it for a video ABM campaign back in 2018 (https://www.ssh-creative.com/abm). One of the most ethical AI tools and services 5/5 https://www.synthesia.io/ethics

Simons-AI-tools-logo_0010_IO25_Gemini_MOD_v33_PART_1_DG_FINAL_54.max-1440x810.png

Google Flow/Veo/Nano Banana, MoonValley, Comfy, Kaiber: Although I was a very early adopter of Runway, but I recently cancelled my subscription. Ethical concerns, not wanting to pay a subscription for what a one-time-purchased plugin can do indefinitely (such as Topaz, SpeedX, Scaleup, etc). AI tools aside, gen-AI video is a much more competitive field now. MoonValley, for example, claims licensed footage, but others lack clear training data details.

Whilst Google AI's have more of a right to be trained on YouTube than Runway has...sort of.

It comes down to more than just how good the generative content is; a key factor is creative control. Some platforms, such as Comfy, enable the video editor or motion designer to control the content and use selective generative AI, not too dissimilar to the AI in Photoshop.

2D ad 3D Character Animaiton

Move.AI: Streamlines motion capture and animation for characters. Ethical Rating: 3/5 - Focuses on motion data, but training transparency is lacking. I have used this on projects where off-the-shelf MoCap, like Mixamo, doesn't have what I need, see https://www.ssh-creative.com/mocap

Wonderstudio: Automates character animation and VFX integration. Ethical Rating: 3/5 - Like Move.AI, its training data practices aren't clear.

I see these as AI tools and not AI content plagiarisers, so I have fewer reservations about using them, and they are very cool.

Reallusion's Cartoon Animator: This powerful 2D animation suite integrates seamlessly with After Effects via a dedicated AE Script, allowing users to export entire projects, including characters, scenes, cameras, and audio, as JSON files that reconstruct layer-by-layer in AE for pro-level VFX, compositing, and motion blur. It's a total game-changer for rigging PSD/SVG assets, applying mocap-driven performances and lip-sync, and polishing in AE without the hassle of manual image sequences. In the past, I've spent hours and hours wrestling with Adobe Character Animator, DuIK, Limber, Joysticks'n'Sliders, and Deekay just to get basic puppetry right—this streamlines it all into a fluid pipeline. Ethical Rating: 3/5 - Relies on user-created assets and licensed content packs, with good privacy controls, but training details for any AI-assisted features (like auto-rigging) aren't fully transparent.

Coding and Development

Replit: A collaborative coding platform with AI-assisted features, great for prototyping interactive media. Ethical Rating: 3/5 - Focuses on code suggestions, reducing IP concerns, but training data details are sparse. Coding is not my kung fu, a little expression writing in After Effects, Zappar, node-based editing in Cinema 4D or script writing in Monday.com are my limitations. So, suddenly being able to potentially create something so far outside of my skill set is quite exciting, even if only to produce a concept prototype before handing over to a professional backend/frontend developer or programmer, which blows my mind a little. I am currently using this to create add-ons for Canva and Express. It can also generate SVG UI elements and toolkits to be edited in Figma.

Code for me: Part of Canva's AI suite, this tool lets me generate interactive code snippets, like buttons, animations, or even simple web elements, from natural language prompts, without needing to dive deep into programming. It's a game-changer for quickly prototyping clickable mockups or embedding dynamic features into presentations and social posts. Ethical Rating: 3/5 - Canva emphasises user privacy and doesn't train on your content by default, but relies on third-party models with limited transparency on sourcing. I have been using this to help me check the code being generated by Replit for Canva.

3D geometry

Simons-AI-tools-logo_0011_Spline.Design.png

Spline.Design: A 3D generator for creating interactive 3D assets, perfect for web and media projects. Ethical Rating: 3/5- Primarily a design tool, but its AI component's training data isn't fully disclosed. I can see this being a lot of fun, however, most of the 3D content I produce tends to be technically and very specific.

Luma AI, NVIDIA's Instant NeRF: I'm excited about these image-to-3D geometry generators for revolutionising 3D asset creation. Ethical Rating: 2/5 - Promising, but training data details are limited.

Adobe Project Neo (Beta): I'm excited about this web-based 3D design tool, which lets me create and edit 3D models with intuitive controls familiar to 2D designers. Its integration with Adobe Illustrator allows seamless import/export of SVG files, making it easy to turn 2D designs into 3D or refine 3D assets as vectors. Perfect for brand graphics and isometric illustrations in my media projects. Ethical Rating: 4/5 - Built on Adobe's licensed data framework, similar to Firefly, though full transparency on its AI components is still evolving. Adobe asked me not to show their logo, which is a shame, as they have the best review on this page and have been assigned the best ethical scores.

Ideation and Prototyping

Adobe Firefly: Great for brainstorming visuals and concept art, integrated with Creative Cloud. Ethical Rating: 4/5 - Uses licensed data and Content Authenticity Initiative metadata, though Midjourney image use is a concern. Regardless, this is usually my first stop when it comes to image gen-AI. Adobe asked me not to show their logo, which is a shame, as they have the best review on this page and have been assigned the best ethical scores.

DALL-E 3: Good for creative visuals, though less photorealistic. Ethical Rating: 2/5 - Faces legal issues over web-scraped data and potential copyright infringement.

Midjourney: Sparks creativity, but its stylised outputs often need ethical tweaks to avoid mimicking protected styles. Ethical Rating: 1/5 - Relies on web-scraped data, leading to lawsuits over IP theft.

Ethics-based, Midjourney creates amazing results. However, both platforms have a terrible onboarding and UI, not very inclusive to non-geeks. Feels like MS-DOS vs. Mac OS. When I use either, it is purely for ideation or a starting point in my creative process, never for final commercial work.

PromeAI turns photos and graphics into consistent sketched images, saving me from manual work in ProCreate (though I miss iPad sketching!). When I am storyboarding, I will often take these images into ProCreate to add a little more human flair (mess) and greater customisation. Ethical Rating: 2/5—Training data isn't well-documented, raising potential IP concerns. I will likely replace this with Google Whisk when it becomes available in the EU.

Figma AI Features like First Draft let me prompt full wireframes or high-fidelity prototypes: "e-commerce dashboard with dark mode", and it spits out editable layers using Figma's component libraries. It's a lifesaver for stakeholder reviews, iterating themes or spacing with quick AI tweaks while staying on-brand. I pair it with my design system for consistency. Ethical Rating: 3/5 - Builds on user libraries without scraping external data, but broader AI accuracy relies on Figma's evolving models with some transparency gaps.

Integration system (aggregated AIs)

ComfyUI: I'm a creative tech geek, but even I get overwhelmed by the flood of new tools popping up, each solving similar problems in isolation. It’s like the old PC vs Mac or 3D Max vs Cinema 4D days...zero compatibility, wrecked workflows. ComfyUI a game-changer here. The original and still-leading open-source node-based workflow engine for Stable Diffusion, Flux, and beyond. Drag-and-drop pipelines give motion designers and video editors total control over upscaling, inpainting, model chaining, and custom logic—no black-box cloud mystery. Runs locally on consumer GPUs. Ethical Rating: 5/5 (open-source, data stays on your machine).

Weavy: This is a node-based platform ties together AI models for text, images, video, and 3D into slick, custom pipelines. I can chain tasks like “prompt to image, upscale, add motion” without hopping between apps. For someone juggling tools daily, it’s a lifesaver for scaling campaigns: build once, tweak for variants, and collaborate on shared canvases with pro compositing. It’s early days, but the control’s ace. Ethical Rating: 3/5 - Pushes user-owned workflows and model flexibility to cut lock-in, but combined AI inputs risk bias if unchecked.

Flora: An emerging extension layer built for ComfyUI, introducing intuitive “flora nodes” for organic, style-blending automations. Train once on a character or mood board, then auto-apply consistent looks across frames or variants. Perfect for rapid iteration on branded motion assets. Ethical Rating: 4/5 (inherits ComfyUI’s transparency; user-trained, user-owned).

Ethical Concerns

ComfyUI + Flora with Weavy: The hybrid power combo—run heavy local processing in ComfyUI/Flora, then push boards to Weavy for team collaboration, version control, and cloud sharing. I’ve been testing this for campaign variants: same pipeline, 40% faster iteration, fully on-brand, zero style drift. Ethical Rating: 4/5 (open-source roots + controlled sharing = minimal IP risk).

As AI generation matures, the real power shift is toward customization—training models on your own assets to enforce brand styles, character consistency, or proprietary visuals without starting from scratch each time. This isn't just fine-tuning; it's about owning the output ethically, with minimal data (often 6-20 images) and commercially safe results. I've started experimenting here to avoid generic outputs, but transparency on training compute and data retention is key. Here's a roundup of standout platforms as of late 2025:

Adobe Firefly Custom Models (Coming Soon): Adobe's rolling out individual access to custom training in Firefly, letting you fine-tune on 6-12 images for characters or "slightly more" for tones/styles—all built on their licensed Stock data for zero IP headaches. Expect integration with Photoshop/Illustrator for seamless export; early access waitlist opens November 2025. Perfect for teams scaling on-brand assets. Ethical Rating: 4/5 (commercially safe base, but watch for ecosystem lock-in).
Mago.studio: A video-first powerhouse for style transfer, but its native image training shines for creators stylizing live-action or 3D refs into custom looks (e.g., upload a mood board, train a "colorful drawing" model in minutes). V3.3 update adds direct visual control—no prompts needed. Great for music videos or animated campaigns. Ethical Rating: 3/5 (user-controlled inputs, but training details are evolving).
Leonardo AI: Upload 10-20 reference images to train a fine-tuned model for game assets, concept art, or stylized outputs; costs ~1450 credits and takes 5 minutes. Its prompt generator and editor make it beginner-friendly for iterating styles. Ethical Rating: 3/5 (focuses on user data, but base models have mixed sourcing—opt for their licensed tiers).
Civitai: Community-driven hub for training/sharing LoRAs (low-rank adaptations) on Stable Diffusion/Flux bases; upload your dataset, train via their tools or integrate with ComfyUI, then download for local use. Ideal for NSFW-free, niche fine-tunes like realism portraits. Ethical Rating: 2/5 (open but prone to unchecked uploads; curate datasets carefully).
Replicate (with FluxGym or AIToolkit): Cloud GPU rental for quick LoRA training on models like Flux; pair with Runpod for heavy lifts. Beginner-accessible via APIs, with pre-built workflows for e-commerce product shots. Ethical Rating: 3/5 (pay-per-use transparency, but compute costs add up—great for one-offs).

These tools bridge the gap between ideation (like Firefly prompts) and production, letting you "teach" AI your vision without deep coding. Start small: Train on a character set for a campaign, then chain into Weavy for video. If ethics are your north star, prioritize Adobe or Leonardo for licensed safeguards.

Generative AI is powerful, but it has challenges:

Intellectual Property: Tools like Midjourney and DALL-E face lawsuits for using artists' work without consent or compensation, threatening creative livelihoods.
Transparency: Many tools don't disclose training data sources, undermining trust. Even Adobe's use of some Midjourney images raises questions.
Job Displacement: AI automation risks reducing opportunities for all 'thinking' roles, not just in the creative industry. I am a geek, I love new tech, but even I can see the end-of-days. AI is devaluing hard-earned creative/technical skills. Someone who can order the best things on a menu is not a chef. Whilst the mantra that AI won't replace jobs, but people who can use AI will, is only partially true. A very small number of AI users can replace whole teams of non-AI users = fewer jobs. This saying is also very short-sighted. I do not believe it is taking the longer view into account. The AI-user's days are also numbered.
Bias and Misinformation: Models can perpetuate biases or misinformation if not monitored, as some note with Grok's data control.
Deepfakes: Voice cloning (ElevenLabs) and avatars (Synthesia) risk misuse, like in the Scarlett Johansson case.

I prioritise tools like Adobe Firefly and Project Neo for their licensed data and advocate for ethical AI development to protect creators.

Tools vs Content Generators

I have referenced this a few times, so let's break this down:

Tools: Adobe Gen-AI, Topaz Video AI, Veed, Descript, PromeAI, Spline.Design, Move.AI, Wonderstudio, Replit, and Project Neo enhance specific tasks (e.g., editing, upscaling, coding, motion capture, 3D design) without creating new content from scratch. They refine existing assets or streamline workflows.
Content Generators: Grok, Perplexity AI, ElevenLabs, Artlist, MotionArray, Soundraw, Adobe Firefly, DALL-E 3, Midjourney, Synthesia, Google Flow/Veo, MoonValley, Comfy, Kaiber, Luma AI, and NVIDIA's Instant NeRF create original content (e.g., text, images, videos, music, 3D models) from prompts, often using large datasets.
Difference: Tools augment human work by enhancing processes, while content generators produce new assets, raising bigger ethical questions about data sourcing and IP.

AI has transformed my workflow, blending creativity with efficiency, but it's the human spark that makes the final content sing. What's your favourite AI tool? Let's chat about pushing the boundaries responsibly!