Replay: Creating Realistic AI Images + AI Voice Overs
Quick note— the above voice note is all AI generated. I’ll cover photos and voice-overs here. Don’t miss the AI voice examples that I recorded below.
To get started with creating your AI photo twin, you will need a Higgsfield account. That link is below.
Here is the prompt I settled on for providing the best results for images. This is the prompt I used for Jill’s photo generation.
"Bathed in soft, natural daylight filtering gently from the right, a poised woman with softly styled hair faces the camera, shoulders straight in a meticulously framed close-up at eye level from half a meter away. Her clear ADD COLOR eyes, luminous yet subtly shaded, meet the lens confidently, framed by ultra-realistic textured skin. Her lips are closed in a soft smile. She wears a simple black t-shirt, complemented by delicate gold hoops and a fine chain, all set against matte white walls that blur gently behind, enveloping her in calm neutrality. The composition, rendered through an 85mm f/1.4 perspective, isolates her presence crisply within the shallow depth of field, preserving every detail without artificial gloss or exaggerated effects. The gentle directional shadows contour her features with refined elegance."
Ultimately, you want to create a base image of yourself that you can then use for other things. I’m not sure how much I’ll use the AI generated images other than maybe changing shirts or making subtle changes to images in order to use them in specific marketing. BUT I wanted to introduce you to the process.
Once you have a base image that you love, you will want to begin each new project with that but you have t keep AI grounded so this is a prompt I learned in a course I took that works for that.
GENERAL MASTER PROMPT TEMPLATE - Adjust to your AI Twin and use every time you create a new image with your base image. This will help ground and guide the AI when creating new images.
Use the base image as the primary identity reference. Maintain the same facial structure as the base image: same eyebrow shape, eye spacing, eye shape, nose structure, cheekbone placement, jawline softness, lip shape, proportions, and overall likeness (70–90% similarity depending on remix settings).
Preserve all natural skin texture:
visible pores, micro-texture, subtle tonal variations, natural shine on high points, soft fine lines, realistic mid-tone contrast, and authentic light diffusion.
Replicate the same level of texture and detail found in the base image. Avoid any smoothing or glam retouching.
Camera angle: straight-on or matching the angle of the base image, eye-level perspective unless otherwise specified.
Lighting: match the lighting direction and softness of the base image unless intentionally changed. If changing lighting, preserve original skin detail and facial structure.
Identity anchors: lightly arched brows, almond-shaped eyes, natural lashes, medium-full lips, soft natural jawline, gentle cheek contour, authentic human detail.
Expression: gentle soft smile with natural eye shape and no exaggerated expressions.
Overall aesthetic: photographic realism, natural makeup, softly feminine glow, true-to-life detail, coherent facial proportions, and consistent identity across outputs.
Negative Prompt: distorted face, altered identity, facial warping, changed bone structure, over-smoothing, airbrushed skin, plastic texture, waxy skin, glam retouching, excessive blur, cartoon-like features, exaggerated eyes, reshaped jawline, altered nose, unrealistic smoothness, glitch.
AI Voice Cloning
For this, you are going to need an Eleven Labs account. You can do a lot with the Starter Account, which is $5/month, but the creator account for $11 seems to be the best value.
https://elevenlabs.io/
In the replay you will see the steps to take in order to establish your voice clone. I’m pretty happy with where mine is now and have only done three or 4 recording sessions with it. I’ll probably continue to perfect it.
Below are some use case examples for you!
This is an audio version of a short newsletter I recently sent out to my database. Easy to do and easy to include at the beginning of your newsletter to give people an easy alternative way to consume your content!
I can see a lot of ways to use a property description like this one. The longer the recording, the trickier I found it so I did play around with the script for this one more than the others but still less “takes” so to say than recording it myself.
Short, easy to generate, ready to use in a bunch of ways…..I think neighborhood snippets are one of the best ways to AI voice.
Meeting Notes:
Jan 12, 2026
Summary
Rebecca Green introduced various AI use case scenarios for content generation, including specific headshots, video narration, and converting newsletters to audio, and demonstrated the AI image generation process using Higsfield AI to create and refine images like Jill Dehlin’s headshot. Rebecca Green also introduced 11 Labs for voice cloning and text-to-speech, advising participants on techniques for natural-sounding audio, which participants, including Sean Gutmann and Jill Dehlin, later noted sounded realistic but sometimes too slow. Rebecca Green suggested practical applications for the generated voiceovers, such as market reports and client updates, with Brooke Andersen noting the benefit of easier voiceover generation, and concluded by recommending the 11 Labs tool over the image generation tool, while Dori Glass inquired about discount codes.
Details
Notes Length: Standard
AI Use Case Scenarios for Content Generation Rebecca Green introduced various use case scenarios for AI in content creation, including generating specific headshots from existing media, video narration for listing tours (referencing Brooke Andersen's existing work), and converting newsletters to an audio format. Other suggestions included creating reusable voiceovers for email templates, such as appointment reminders, and generating social media content.
AI Image Generation Process and Results Rebecca Green demonstrated how AI can generate images, revealing that all four example photos shown were AI-generated, based on merging and regenerating two different photos. She explained the process using Higsfield AI, specifying settings like selecting "Nanobanana Pro," changing the orientation to 9:16, and setting the detail to 4K for the highest clarity, noting that the AI automatically smoothed out features like wrinkles and lines.
Headshot Generation with Prompts Rebecca Green used Jill Dehlin’s headshot to demonstrate image generation, emphasizing that starting with a straight-on, basic photo yields the best results. She mentioned using positive and negative prompts, which she gathered from different sources, including a program by Jason Pantana. The regenerated image of Jill Dehlin included a black shirt and gold earrings, based on a prompt Rebecca Green added. Rebecca Green noted that the second generation of Jill Dehlin’s photo was significantly better than the first, demonstrating the need for iteration and tweaking.
Voice Cloning and Text-to-Speech using 11 Labs Rebecca Green introduced 11 Labs for creating voiceovers, noting that the paid subscription, which is around $9 a month, is necessary for cloning one's own voice. She recommended using the "instant voice clone" feature, which requires recording short 10-second audio snippets. Rebecca Green advised against reading from a script generated by ChatGPT, as it tends to sound robotic; instead, she suggested recording while having a natural conversation to capture more natural vocal inflection.
Applying and Refining Text-to-Speech Output Sean Gutmann and Jill Dehlin noted that they could not initially hear the audio examples played by Rebecca Green, who then played them successfully from their phone. Rebecca Green demonstrated text-to-speech conversion, explaining that adding periods creates pauses, which helps control the speed of the audio. They noted that the generated voice sounded realistic but that the speed was sometimes too slow, necessitating adjustments to speed, similarity, and style exaggeration settings. Rebecca Green also mentioned the importance of selecting the appropriate model for the best result, noting that the model choice significantly impacts the output.
Practical Applications and Workflow for Voiceovers Rebecca Green suggested using the text-to-speech feature for market reports and client updates, enabling conversations instead of just sending static data. Brooke Andersen noted that the ability to generate voiceovers easily would be helpful, as their current process for video tours often requires multiple takes. Rebecca Green suggested using BombBomb to embed audio into emails, though they clarified that the audio would still open on a different platform due to email platform limitations. Rebecca Green plans to create and share sample voiceover pieces, including a market report and a thank-you note for listing appointments, demonstrating the "wash, rinse, repeat" nature of reusing content.
Recommendations and Cost Consideration Rebecca Green stated that if one were to choose between the two AI tools, they would recommend the text-to-speech tool (11 Labs) over the image generation tool. Dori Glass asked about discount codes, but Rebecca Green confirmed they did not have any for 11 Labs or Nano Banana at the time. Rebecca Green cautioned attendees to only subscribe if they plan to use the service, emphasizing that small monthly fees can accumulate to significant annual expenses.
