0:00 / 0:00

Comment “AI” to try out this new feature that will allow you to create ultra realistic images in just a couple of clicks 🤩 @higgsfield.ai just launched Soul 2 and it’s absolutely crazy , try it and you’ll understand what I mean 😍 #higgsfield #soul2 #higgsfieldsoul2 #higgsfieldpartner

Simone Ferretti

@sferro21

INSTAGRAM · 2026-02-20Source

3likes

528comments

Remix This

Recreate with Kling 3

Make your own AI viral video

Prompt

GLOBAL LOCK: A vertical 9:16 creator-marketing Reel, approximately 33 seconds, built around one recurring host and a dark-mode AI character-generation interface. Keep three visual layers consistent across the whole video: (1) the host, a white male in his late 20s to early 30s with side-parted brown hair, slim build, expressive face, clean-shaven, wearing a fitted off-white knit sweater and speaking into a matte-black desktop microphone, lit by a warm amber key and soft vignetted studio background; (2) stylized portrait outputs of the same handsome male AI character, usually white, early 20s to early 30s, chiseled jaw, thick dark hair, slim-athletic build, shown in different fashion/editorial presets such as city streetwear, convenience-store candid, studio portrait, tank-top fashion, foggy road noir, cowboy desert, and black-and-white urban scenes; (3) Higgsfield.ai interface captures in dark mode featuring the Character section, Higgsfield Soul 2.0 highlighted in the left model list, a grid of example source faces, preset tiles labeled Editorials, Fashion, Street Photography, Double exposure, a bright lime-green Generate button with a coin cost indicator, and an Animate button on selected outputs. The pacing must stay aggressive and social-native with a new visual beat every one to two seconds, strong contrast between warm host footage and colder generated sample cards, crisp UI sharpness, black/charcoal backgrounds, neon-lime accent labels, and one energetic male speaker throughout with close-mic, dry, high-intelligibility audio. Lips are visible during all host sections and sync must feel tight.

[00:00-00:03] Start on a dark background with bold white uppercase text reading STOP DOING THIS, flanked by red X marks. Under the headline, show generic AI male portrait samples: first a black-coat city street shot, then a casual black sweater portrait, then another generic urban fashion image. The host appears in a rounded rectangle at the bottom, urgently raising one hand toward camera as if interrupting the viewer. Audio: same male host delivers a sharp pattern-break hook telling viewers to stop making the same boring AI character photos.

[00:03-00:07] Cut between the host in warm studio close-up and more bland sample outputs: a crouched white-sweater studio pose, a convenience-store fashion portrait with bomber jacket and bow tie, another convenience-store variation. The host points upward with both index fingers while speaking quickly. Camera on the host remains static medium close-up with 35mm to 50mm lens feel, shallow depth, warm amber falloff. Audio: one speaker, emphatic, corrective tone, lips fully visible.

[00:07-00:11] Introduce stronger preset-driven examples. Show a clean editorial portrait card labeled Editorials, then a Fashion preset with a white ribbed tank top, then Street Photography over a bright outdoor male portrait, then Double exposure with a grayscale silhouette overlay. Each sample occupies the upper two-thirds while the host continues in the lower panel. The transition rhythm should feel like flipping through creative options rather than a tutorial menu. Audio: host pivots from criticism to the better alternative.

[00:11-00:14] Briefly isolate the Higgsfield.ai logo on a dark bar, then cut to the platform interface. Show the Character tab area with Soul 2.0 in the model list highlighted and the host below continuing to explain. Use dark graphite UI, lime-green badges, and readable white text. Audio: same speaker names the tool and frames it as an easier route to ultra-realistic character creation.

[00:14-00:18] Show a grid of source reference portraits inside the character workflow: multiple male selfies and studio shots, the cursor hovering over them as if choosing a base identity. Host remains bottom-center, speaking calmly but with momentum. Emphasize that one character identity can be turned into many outputs. Audio: host explains consistency and customization, crisp consonants, no background reverb.

[00:18-00:21] Cut to a full-height preset card of a standing male figure against a white seamless with a lime Presets label, then to the generation composer showing a dark prompt box, a character token or preset mention, and a lime Generate button with a coin cost. Cursor movement should imply that generation is about to happen. Audio: host explains that the system can create polished images in a couple of clicks.

[00:21-00:24] Reveal generated outputs in different environments: a dark cinematic portrait of a bespectacled man, a convenience-store streetwear shot with Presets badge, and an outdoor coastal portrait with Animate highlighted in lime. The host gestures with one hand as if listing options. Color shifts between cool storefront daylight, neutral portrait lighting, and warm natural outdoor scenes while the UI frame stays dark.

[00:24-00:28] Expand the sample range further with a foggy road full-body shot in a long black coat, a desert cowboy standing in front of a stepped stone structure, and a top-down tank-top fashion portrait. These three outputs should feel dramatically different in location and styling while keeping premium realism and the same polished character aesthetic. Audio: same male narrator sells variety, speed, and realism for creators.

[00:28-00:31] Tighten into darker cinematic portraits: a serious close-up male face against a charcoal backdrop, then a black-and-white street portrait with overlaid CTA text Comment "AI", then a fashion portrait with the same CTA treatment. Keep typography large, bold, white, and lime-yellow, centered over the images. The host points upward from the bottom frame to reinforce the CTA timing.

[00:31-00:33] End on another fast CTA repetition using the strongest portrait samples while the host lands the final line. Maintain the warm studio box below, sharp microphone silhouette, and dark premium brand palette. Audio: one male speaker, punchy final comment-gate instruction, no fade, no music swell overpowering the words.

NEGATIVE PROMPT: avoid identity drift between generated male portraits, avoid uncanny skin texture, avoid distorted eyes or asymmetrical jawlines, avoid over-smoothed plastic faces, avoid broken hands in host gestures, avoid unreadable UI labels, avoid cluttered text overlays beyond STOP DOING THIS and Comment "AI", avoid fake logos, avoid low-resolution preset cards, avoid inconsistent sweater color on the host, avoid muddy shadows on the warm studio shot, avoid robotic speech, lip-sync mismatch, clipped peaks, harsh sibilance, or over-compressed voice.

How sferro21 Made This Higgsfield Soul 2 Character Generator AI Video - and How to Recreate It

This Reel is a smart AI creator ad disguised as a fast aesthetic intervention. Simone Ferretti opens with a bold correction, "Stop doing this," over generic AI male portraits that already look familiar to anyone who has spent time in AI image communities. He immediately contrasts those bland outputs with warmer, more premium-looking character results generated inside Higgsfield.ai using Soul 2.0. The visual vocabulary is consistent and easy to remember: dark charcoal background, red X marks, bright white text, a warm talking-head setup with a black microphone, lime-highlighted UI controls, preset labels like Editorials, Fashion, Street Photography, and Double exposure, plus a green Generate button and an Animate button shown inside the workflow. The piece works because it sells transformation, not software. It says: stop making generic AI men, start making a reusable character system that can shift from convenience-store candid to fashion portrait to noir road scene to cowboy editorial in a few clicks. For indie creators, that is a strong promise because it touches avatar building, ad creative, social content, and identity consistency all at once.

What You're Seeing

The opening is a direct attack on stale AI aesthetics

The phrase STOP DOING THIS is not just clickbait. It is attached to a very specific type of output: polished but generic AI male portraits that feel overfamiliar. That matters because the Reel begins by naming the problem visually before pitching the solution.

The host keeps the message human and fast

The talking-head setup is simple but strong: warm background, soft key light, off-white sweater, and a black microphone dead center. This gives the Reel a creator-native feel instead of a polished SaaS ad feel, which makes the recommendation easier to trust.

The generated examples are grouped by style logic

The Reel cycles through examples that clearly belong to different visual buckets: editorial portrait, fashion portrait, street photography, double exposure, convenience-store candid, cinematic headshot, foggy long-coat shot, cowboy frame, and outdoor coastal portrait. That variety is the feature proof.

The interface shots are brief but strategically placed

You see Higgsfield.ai, the Character workflow, the Soul 2.0 model entry, sample character references, preset cards, and the Generate button. None of these shots stay on screen long, but together they provide just enough proof that the output is attached to a real workflow.

The CTA closes the loop using the best-performing images

At the end, Comment "AI" is layered on top of the strongest portrait outputs rather than on a blank end card. That keeps the visual promise alive right up to the conversion moment.

Shot-by-Shot Breakdown

Time range	Visual content	Shot language	Lighting & color tone	Viewer intent
0:00-0:03 (estimated)	STOP DOING THIS headline over generic AI male portraits	Static card-style images with quick swaps, host panel below	Dark background, white type, red X accents	Pattern interrupt and call out a common mistake
0:03-0:07 (estimated)	Host continues speaking while more generic portraits flash	Talking-head plus sample-image alternation	Warm studio lighting against darker top-frame content	Build agreement before introducing the solution
0:07-0:11 (estimated)	Preset-driven examples: Editorials, Fashion, Street Photography, Double exposure	Quick showcase cards, one image per concept	Mixed looks, from studio clean to outdoor natural light	Show obvious stylistic range fast
0:11-0:14 (estimated)	Higgsfield branding and Soul 2.0 workflow entry	UI insert with host composited underneath	Dark-mode UI with lime-green highlights	Prove the examples come from a specific tool
0:14-0:18 (estimated)	Character reference grid and source identities	Cursor-guided screen recording	Neutral UI blacks with bright image thumbnails	Explain character consistency without over-teaching
0:18-0:21 (estimated)	Preset card and Generate button in the composer	Product proof shot with interface zoom	Dark controls, lime CTA button	Make the workflow feel easy and clickable
0:21-0:28 (estimated)	Series of generated portraits in different environments and styles	Portfolio-like sample carousel	From cool convenience-store daylight to foggy noir to desert sun	Expand imagination and keep retention high
0:28-0:33 (estimated)	Comment "AI" over top-performing portraits while host finishes pitch	Repeated CTA overlays on strong hero images	Dark premium portrait palette with bright text accents	Convert curiosity into comments without losing aesthetic value

Why It Went Viral

It starts with criticism, which is naturally scroll-stopping

Many AI tool videos begin with praise. This one begins with rejection. By saying STOP DOING THIS and pairing it with familiar-looking generic outputs, the Reel invites viewers to self-diagnose whether their own AI images feel stale. That is a much stronger opening than a generic product announcement.

It turns a tool feature into a status upgrade

The promise is not simply "make images faster." The promise is "stop looking amateur." That shift matters because creators respond strongly to signals of taste, distinctiveness, and premium-looking outputs. The examples here are curated to look better, not just different.

The style variety is broad enough to unlock multiple audiences

Fashion creators can focus on the tank-top and editorial portraits. Personal branding creators can focus on the studio headshots. Story-driven creators can focus on the foggy road and cowboy scenes. The Reel is effective because one workflow appears to serve many niches.

The host's tone is corrective but not preachy

He sounds like a creator sharing a shortcut, not a corporate trainer reading a feature list. That tone fits Instagram well, especially when the content is educational but still meant to feel fast and stylish.

The CTA is friction-light and reward-clear

Comment-gating works here because the exchange is obvious. The viewer is not being asked to remember a URL or navigate away. They just comment AI if they want access, which is a low-effort action with a direct reward.

Platform Signals

The 0-3 second hook is visually legible with the sound off

The big white text, red X marks, and overfamiliar portrait examples make the opening understandable instantly. That is useful on Instagram because many users decide whether to stay before they fully process the spoken line.

The edit keeps changing information type

The Reel rotates between critique, host, example outputs, UI proof, and CTA. That variety is a retention advantage because the viewer never sits inside one visual mode for too long.

The content has high save potential

Creators can save this for three different reasons: better prompt taste, better avatar/character workflow ideas, and a specific tool to test later. That makes the post useful beyond the first view.

5 Testable Viral Hypotheses

Hypothesis 1: Negative framing lifted the hook rate

Observed evidence: the Reel opens with STOP DOING THIS instead of a feature celebration. Mechanism: criticism triggers curiosity and self-comparison faster than praise. Replication: test a corrective hook versus a neutral "new feature" hook.

Hypothesis 2: The examples were varied enough to keep viewers watching

Observed evidence: the Reel shows convenience-store, fashion, editorial, fog, desert, monochrome, and outdoor portrait looks. Mechanism: viewers stay longer when each new card suggests another possible use case. Replication: stack at least five visibly different outputs in your demo.

Hypothesis 3: The host boosted credibility more than a full-screen screen recording would

Observed evidence: the host remains visible through much of the video. Mechanism: a face can carry trust and pacing while the top half carries proof. Replication: keep your explainer visible during the most important proof shots.

Hypothesis 4: Soul 2.0 felt like a system, not a one-off prompt

Observed evidence: the Reel shows source identities, presets, generation controls, and animate options. Mechanism: viewers perceive a repeatable workflow instead of a lucky result. Replication: show one or two workflow screens, not just final outputs.

Hypothesis 5: Comment-gating increased visible engagement without hurting clarity

Observed evidence: the final CTA is layered over the strongest images and repeated. Mechanism: the action is simple and timed right when the viewer has already seen enough proof. Replication: ask for one short keyword comment instead of a vague engagement prompt.

How to Recreate It

Step 1: Start with a mistake your audience recognizes

Do not begin with the tool. Begin with a bad output pattern, a tired aesthetic, or a workflow shortcut your audience knows is real.

Step 2: Build one warm host setup

A simple talking-head setup with one warm light, one microphone, and a dark vignetted background is enough. The original Reel proves you do not need a large set.

Step 3: Prepare one consistent character identity

If you are generating AI portraits, keep a character sheet with 4 to 8 identity references, hairstyle notes, body type notes, and outfit anchors. This is what makes range feel intentional rather than random.

Step 4: Organize presets by visual use case

Do not present random outputs. Group them into clear buckets like editorial, fashion, street, cinematic, noir, and lifestyle.

Step 5: Show just enough interface to remove doubt

Include the model name, the character workflow, and the generate state. That gives the viewer confidence that the results are reproducible.

Step 6: Use output contrast as your pacing engine

Move from weak/generic examples to stronger/premium examples. The contrast is what makes the tool feel transformative.

Step 7: Write short spoken beats

Your script should move through five functions quickly: call out the problem, name the tool, show the workflow, show the range, give the CTA.

Step 8: End on your strongest three images

Do not put the CTA on your weakest frame. Put it on the portraits that make people want the workflow.

Step 9: Match the CTA reward to the curiosity gap

If you want comments, the reward must be specific: access, link, prompt, breakdown, or tutorial. In this case, "comment AI" clearly promises tool access.

Growth Playbook

3 opening hook lines

1. Stop making AI portraits that all look exactly the same.

2. If your AI characters still look generic, this is the fix.

3. This is the fastest way I have found to make one character feel actually premium.

4 caption templates

Template 1: Most AI character posts still look way too generic, and that is why they do not stick. This workflow gives you cleaner identity control and better style range fast. Want to try it? Comment AI.

Template 2: If you are building AI influencers, ad creatives, or branded characters, this is the kind of tool shift that matters. The presets are not the point, the repeatability is. Comment AI for the link.

Template 3: I like tools that turn taste into a workflow, not just into one lucky image. Soul 2.0 feels useful because you can move from editorial to street to cinematic fast. Want the setup? Comment AI.

Template 4: This is a much better way to create ultra-realistic AI characters than posting the same basic portrait over and over. Use one identity, build more range, and create stronger creative assets. Comment AI.

Hashtag strategy

Broad: #AIImage #GenAI #AIContent. These help catch general AI creator interest.

Mid-tier: #AICharacter #AIInfluencer #AICreativeWorkflow #AIPortrait. These align with the actual use case shown in the Reel.

Niche long-tail: #Higgsfield #Soul2 #HiggsfieldSoul2 #CharacterGeneration #AICharacterDesign. These target viewers already searching for this exact style of workflow.

FAQ

Why does this Reel hook so fast?

Because it starts by rejecting a common bad output pattern instead of slowly introducing the tool.

What is the most important visual decision here?

Showing clearly different portrait styles while keeping the character quality premium and believable.

Why does the host stay on screen so often?

The host makes the workflow feel recommended by a person, not just displayed by software.

How can I keep my AI character from looking generic?

Use a stronger identity base, then push style variety through presets, environments, and wardrobe changes instead of minor face edits only.

What makes this different from a simple before-and-after demo?

It shows a repeatable system with references, presets, generation UI, and multiple outcome types.

Why is the comment CTA effective here?

It is short, low-friction, and attached to a clear reward after enough proof has already been shown.

Would this format work for female AI characters or brand mascots too?

Yes, as long as you keep the same structure of critique, proof, range, and friction-light CTA.